This PDF file contains the front matter associated with SPIE Proceedings Volume 13090, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

The icing will affect the characteristics of the radome with “High penetration in the band, strong stealth out of band”. Most of the existing research is on the electromagnetic properties calculation of frequency selective surface (FSS) in ideal environments, and there is a lack of research on the electromagnetic properties calculation in icing environment. In this paper, a new algorithm is proposed, and taking icing monopole FSS as the research object, the hybrid algorithm is verified, which has the advantages of improving the calculation speed and accuracy. On this basis, the influence of icing, icing thickness and ε(ε) of the ice layer on the wave-transmitting characteristics of FSS is analyzed. The calculation results show that: Icing will make the resonant frequency of FSS smaller; and as the icing thickness increases, the number of frequency resonance points increases; as the ε of the ice layer increases, the resonant frequency of the FSS decreases. It can provide an important theoretical basis for the application of FSS in the icing environment.

Cyclone multiphase flow in the annular space is widely used in fluid machinery, such as burner and pneumatic conveying. However, the annular flow field is complex, and the related research is not sufficient. To improve the safety and efficiency of equipment, this paper proposes a method for detecting the motion state of swirling fluid in annular space by integrating computational fluid dynamics (CFD) and electrical capacitance tomography (ECT), calculates the motion characteristics of swirling multiphase flow in the annular space using the CFD, and visually measures the distribution and motion state of swirling multiphase flow in the annular space using the ECT. Numerical simulation and experimental results show that the results of the two methods are in good agreement, indicating that the model selected in this paper in the CFD is correct. The CFD effectively reveals the distribution of swirling multiphase flow in the annular pipe, and the ECT can accurately reconstruct the position and size of swirling multiphase flow in the annular space. The combination of these two methods provides a new idea for the study of multiphase flow in annular space.

This paper presents a non-contact method for characterizing junction temperature (Tj) of GaN-based blue LED using the centroid wavelength combined with the full width at half maximum (FWHM). By establishing the relationship between the centroid wavelength, FWHM, junction temperature, and driving current of LED, a method employing combination centroid wavelength and FWHM for junction temperature characterization is derived. Ten GaN-based blue LEDs are chosen as samples, which junction temperatures are measured under various operating conditions using both the new method and forward voltage method. The research demonstrates that the new method characterizes effectively Tj of GaN-based blue LED with an average error of 2.3°C. This study holds significant importance in LED design and application.

The prediction of skarn type polymetallic ore is very important. This paper takes a certain skarn type polymetallic ore in Mongolia as the research object. Through high-precision magnetic survey on the ground, the plane position of the ore-forming favorable area is quickly optimized. The analytic signal amplitude of No. It is found that the skarn contact zone in the middle of No. 3 anomaly area is relatively developed, with a thickness of 70-350m. Combined with the results of Eulerian inversion of high magnetic data, it is considered that the shallow 200m is a favorable location for mineralization. Finally, the 2.5D human-computer interactive inversion of magnetic anomaly is used to predict the shape and scale of the ore body. The drilling data of later construction is roughly consistent with the position and thickness of the ore body. It has been proved by practice that good prospecting effect has been achieved in the exploration of this type of skarn type polymetallic ore by using audio frequency magnetotelluric survey and high-precision magnetic survey on the ground.

The oxidation reaction of O2 adsorbed at different sites of Ti (010) surface was simulated by density functional theory. The most common sites of Top, LB and SB were selected to study the surface adsorption and reaction of oxygen molecules, and the adsorption energy, geometric structure and electronic structure of all sites were analyzed. SB site is the most stable adsorption site. There is hybridization between the 2s and 2p orbitals of O2 molecule and the 4s/3p/3d orbitals of Ti atom, which will lead to the energy reduction of Ti (010) surface, and eventually lead to the transformation of surface structure and crystal type The adsorption at the LB/SB sites are more likely to lead to the hybridization of the O2 molecule and the electron orbitals of the superlayer and sublayer of the titanium substrates than that at the Top site.

The infinite time evolving block decimation in the presentation of infinite matrix product states based on machine learning is adopted to research a two-leg spin diagonal coupled setback ladder model. In order to deeply research the topological order and quantum phase transition of the model, odd string correlation and even string correlation are introduced, and the fidelity, local order parameters, odd string correlation and even string correlation order parameters are calculated. String correlations exhibit different behaviors. Based on the characteristics of the string correlation, the phase diagram of the ground state sequence parameters of the two-leg spin diagonal coupled setback ladder model is given. In addition, the results of phase transition point, continuous phase transition and order parameter detected by fidelity are consistent.

This study will validate the effectiveness of the derived mathematical formulas correlating the normal and tangential components, a_n and a_t, of acceleration with its rectangular components, a_x and a_y, by applying the derived formulas in solving some problems in engineering mechanics particularly in curvilinear motions.

This paper presents a simpler method of evaluating integrals of powers of natural logarithm. A reduction formula is derived to evaluate the integral of the nth power of natural logarithm. The reduction formula is then repeatedly applied until a generalized formula is derived. The exponents and coefficients of the derived formula showed very recognizable patterns. The exponents of the logarithmic terms are in decreasing order while the coefficient of each term is a permutation.

Manipulating a mathematical equation is usually one of the common problems of students enrolled in mathematics course. For some, Mathematics is one of the most difficult courses. Contrary to this belief, Mathematics is a fun and enjoyable course if solving equations could be done in a short but understandable method. This paper developed, a fast and reliable formula that would simplify sin⁶nθ and cos⁶nθ using Half Angle Identities. This formula offers a simplified approach in simplifying sine and cosine function with exponent 6 to a function of cosine with exponent 1 with any value of n. From the algorithm of solving half angle identities the new formula will emerge.

One of the major topics in Integral Calculus is evaluating integrals of powers of trigonometric functions. Conventional methods use trigonometric identities to transform integrands into forms where integration formulas can be directly applied. These methods are usually long, tedious, and time consuming. Since many students are not patient enough to deal with lengthy and complicated solutions, the objective of this study is to derive direct integration formulas in evaluating integrals of powers of tangent, and to outline techniques that will simplify the solution. Expository research method was used by first deriving a reduction formula. The reduction formula was applied repeatedly to the integral of the nth power tangent until generalized equations and direct integration formulas for odd and even powers were derived. Since the derivation process involved recursive relations, the coefficients and exponents of the derived generalized equations showed certain patterns and sequences that were used as the basis in outlining simplified techniques. The derived formulas and simplified techniques can speed up problem solving since it can generate “one-liner” solutions in contrast to the long and tedious process using the conventional methods. The simplified techniques can also be used as useful inputs to computer programs (systems application process) that can generate mathematical solutions.

Optimization is a method of determining the optimal value of an objective function while satisfying the given restrictions. This study used the method of Lagrange multipliers to set up an auxiliary function to determine the optimal dimensions that will minimize the total surface area of a right pyramid with regular n-gon base subject to a fixed volume. The derived formulas can be utilized by the companies to lessen their packaging expenses.

Ultraviolet (UV) photodetectors have significantly impacted people's lives in UV communication, flame monitoring, missile plume detection, and astronomical studies. Graphene (Gr)/gallium nitride (GaN) Schottky junctions (SJs) have a more promising prospect in UV photoelectric devices than metal/GaN SJs owing to their higher efficiency, wider photoactive area, and better thermal stability. However, previous studies have neglected the impact of GaN polarity on photoelectric characteristics. In this study, the photoelectric characteristics of a Gr/GaN SJ were simulated using Silvaco TCAD, and the effect of the GaN polarity on the detector performance was investigated. The photodetector with a Gr/GaN SJ achieved a responsivity of 0.26 A/W, response time of 0.1 ns, and external quantum efficiency of 89.8% at a reverse bias of -0.5 V. Additionally, the on/off ratio and detectivity of the photodetector with a Gr/Ga-polar GaN SJ were approximately 10⁶ and 10³ times those of the photodetector with a Gr/N-polar GaN SJ, respectively. These considerable disparities are attributable to the higher Schottky barrier and the substantial built-in electric field between Gr and Gapolar GaN. The results provide theoretical guidance for preparing GaN-based SJs and their application in photoelectric devices.

In this work, for a parabolic inverse problem with control parameters, a space-time spectral penalty method based on a Legendre-Galerkin-Chebyshev collocation in space, and a Legendre-tau method in time is presented. According to the scheme of space-time spectral penalty method, the convergence analysis is carried out, and the approximation results of the Chebyshev interpolation operator in the Legendre norm are given. Suitable basis functions are used in each direction, which yields an algebraic system of sparse matrices. Numerical examples are compared with some other methods to confirm the efficiency and capability of the proposed method.

This paper undertakes theoretical analysis and numerical simulation investigation on the axial force of the V-shaped clamp in order to improve its sealing effectiveness. First, the key affecting parameters of axial force, such as bolt preload and V-shaped contour angle, were determined using force analysis. The influence law of axial force was then discovered using numerical simulation. The research findings demonstrate that the bigger the bolt preload, the greater the axial force, but the greater the axial force fluctuation; the smaller the V-shaped contour angle, the greater and more uniform the axial force. V-shaped clamps with reasonable size designs and assembly processes can effectively boost their axial force.

In order to carry out high-order applications such as feature extraction, registration, surface reconstruction, and visualization, it is necessary to separate ground and non-ground points in point cloud data, and a point cloud filtering algorithm based on lidar is proposed. In this paper, three typical point cloud filtering algorithms (Point Cloud Filtering Algorithm of Moving Surface Fitting, Progressive Triangulated Irregular Network Densification, Mathematical Morphology Point Cloud Filtering Algorithm) are summarized from the research status at home and abroad, the specific process of the algorithm and the evaluation of filtering effect, and the problems in practical application are solved by comparing the characteristics of the filtering algorithms, and the follow-up research direction is analyzed. Overall, for flat and simple terrain, the three typical point cloud filtering algorithms mentioned above have good filtering effects, but for discontinuous and complex terrain, the filtering effect is poor, especially for the point cloud filtering algorithm of moving surface fitting with poor adaptability to discontinuous terrain. PTD operation is complex and the filtering efficiency is low. Mathematical morphology point cloud filtering algorithm is mature in theory and widely used. At present, the point cloud filtering algorithm of lidar has been developed more maturely, but there is still much room for improvement in realizing high efficiency and adaptive filtering.

In order to optimize the method and technology for the rapid determination of copper ore grade using Energy Dispersive X-ray Fluorescence (EDXRF), improve the accuracy of EDXRF analysis, guide the development and utilization of copper ore resources, and enhance production efficiency, the optimal operating conditions for the EDXRF analyzer were determined using Monte Carlo (MC) simulation and physical experiments. The recommended settings were a tube voltage of 40 kV and a tube current of 10 μA. Under these optimal conditions, the copper (Cu) content in various types of copper ore samples was analyzed. The copper ore was categorized into geological samples (including raw ore samples and tailings samples), copper concentrates, and matte. The correction of matrix effects was carried out using a binary regression method. Experimental results showed that the average absolute error of analyzing geological samples by EDXRF was 0.035%, with an average relative error of 4.08%. The average absolute error for analyzing copper concentrates was 0.238%, with an average relative error of 1.01%. For matte analysis, the average absolute error was 0.421%, with an average relative error of 1.01%. The accuracy and precision of the method meet the production requirements of enterprises, and it holds promising prospects for widespread application.

The RANS equation and BSL k-ω turbulence model are employed to solve the internal wave generated by a moving submerged body in a continuously stratified fluid, utilizing unsteady methods. Consequently, the relationship between internal wave patterns and diving depth as well as the maximum crest and trough height with respect to depth can be determined.

In the actual industrial production, the monitoring and visual measurement of the temperature distribution in the boiler furnace play an important role in the safe operation of the boiler. Acoustic tomography has the advantages of contactless sensing, and is widely used in industrial furnace temperature field detection and other related fields. The main technique is the temperature field reconstruction model and algorithm. In this paper, for the integral on the line segment located in a small area, instead of the function value of the line segment formula instead and using the rectangular formula, but the function value on the Gaussian node into the linear combination of the region and its vicinity and using the Gaussian quadrature formula for calculation. In order to solve the uncomfortable qualitative problem caused by the many few variables, we construct the relationship between two directional derivatives and partial derivatives. Numerical simulation experiments show that the established model can reconstruct the temperature distribution very well.

One effective solution to address the growing issue of electromagnetic pollution is the creation of high performance electromagnetic absorption materials. Herein, 3D net-like composites of Fe₃O₄/MWCNTs were fabricated using the facial hydrothermal procedure. Benefiting from the synergistic effect between dielectric and magnetic loss, multiple reflections and scatterings, Fe₃O₄/MWCNTs composites exhibiting efficient electromagnetic absorption performances. Remarkably, at 5.64 GHz, the maximum reflection loss (RL) can reach as high as -45.70 dB, while the wide effective absorption bandwidth could cover 7.96 GHz (3.72-8.16 GHz, 13.88-17.40 GHz) with the thickness range of 3.1-4.7 mm. Overall, the fabricated 3D net-like Fe₃O₄/MWCNTs composites are expected to be powerful candidates for light weight, high performance and wide bandwidth electromagnetic absorption materials.

In the previous experiments, the motion of the free-falling bodies usually did not take into account the impact of other external resistance, such as the drag of the medium, which is mainly to highlight the meaning of Earth's gravity because it is undeniable that the significance of the Earth gravity causes quite a lot of changes to the free fall movements. However, resistance is also one of the most indispensable factors which affect the motion of the free-falling bodies. The resistance may affect the speed, the direction, and the autorotation rate of the object. In the field of engineering, we should not only consider the type of various resistances of the media but also the things that change the size of the resistance, for instance, the humidity and the temperature. As a result, we propose a falling model that takes into account changes in medium resistance in this thesis.

Numerical simulations of a radial swirl burner were performed using the RNG k-ε turbulence model. The simulated results were compared with experimental data under both pure flow and combustion conditions to analyze the swirling flow and flame combustion characteristics of the non-premixed burner. The Research result illustrated that within the swirl, there exists a central recirculation zone (CRZ), and an increase in the width of the CRZ would affect the radial expansion of the swirl jet (SJ). Methane combustion primarily occurs within the shear layers (ISL and OSL), resulting in a conical flame shape. Additionally, a flame ring forms near the injection orifice and the recirculation bubbles generated by the swirl generator contribute to the stable burning of the flame ring.

Tilting tile journal bearings are widely used in rotary condensers. The oil film flow in them gradually gets rid of laminar flow and develops into turbulent flow. To accurately analyze this kind of superlaminar flow flow process, it is necessary to correctly describe and simulate the superlaminar flow flow process. In this paper, direct numerical simulation is used to study the superlaminar flow and find that the superlaminar flow is near-wall turbulence. Aiming at the near-wall turbulence with a low Reynolds number, the SST model modified by a low Reynolds number is used for simulation. The simulation results are compared with the experimental and direct numerical simulation results. The results show that: The SST model modified by low Reynolds number can accurately predict the transition from laminar flow to turbulent flow, and verifies that the phenomenon of superlaminar flow in tilting journal bearing is Taylor-Couette near-wall turbulence.

This study utilizes data acquired from the CosmicWatch (CW) apparatus, encompassing metrics such as pressure, temperature, dead time, and time stamp, to conduct an exhaustive analysis of their respective impacts on muon coincidence events and the apparatus's inherent functionality. The investigation extends to the assessment and rectification of discrepancies inherent in the dataset, employing methods of estimation, while ensuring data fidelity through Gaussian distribution evaluation. Additionally, this study employs various parameters within a consistent methodology to observe the environmental effects on muon detection outcomes.

Origami has always been regarded as a folk art, and it has gained attention in recent years, from the passing of experience from generation to generation to the theoretical formation and development of a discipline. To give a clearer description of the motion of some cylindrical origami mechanisms, in this paper, some classical models are analysed kinematically by using common methods in mathematics, and some special properties in cylindrical origami are proved. Finally, the properties are analysed by the relation between the rotation angle between the elements and one of the connecting rods. The mathematical method used in this paper is ingenious and simple, which reduces the difficulty of analysing origami and opens up a possibility for simplifying the analysis of origami.

Liu [1] has already been proven a compact embedding from fractional Sobolev spaces HW^s,p (Ω) to L^q (Ω) , where Ω⊂Hⁿ is a bounded extension domain and q∈[1, p] . In this paper, we prove a compact embedding from fractional spaces HW^s,p(Ω) to L^q(Ω) under sp < Q , where Ω⊂Hⁿ is a bounded extension domain and q ∈ [1, p*). Moreover, we give a Sobolev embedding for fractional spaces in the Heisenberg group.

This paper explores the redefinition of operational quantities in radiation protection, analyzing their impact on external radiation monitoring. It delves into the limitations of current operational quantities defined by ICRU Report 39, such as their inability to accurately measure certain types of radiation and energy levels. The new definitions proposed in ICRU Report 95 are discussed, focusing on how they offer a more accurate representation of radiation risks and improve the effectiveness of radiation monitoring instruments. The paper also examines the future development of operational quantities, suggesting individual customization and the application of machine learning for optimization. The redefined operational quantities aim to enhance radiation protection by providing more precise and practical measures for various radiation types and environments.

This study aims to establish a large-volume extrapolation ionization chamber with water-equivalent materials for the absolute measurement of the absorbed dose in 1 cm depth water by I-125 particle source based on ionization methods. RW-1, RW-3, PMMA, and Blue Water, were selected for investigation, taking into account internationally adopted water-equivalent materials and commonly used alternatives. MCNP5 and EGSnrc software were used to developed models for the 6711 particle source and extrapolation ionization chamber. Varying water-equivalent materials, comparative analysis with WATER was conducted in terms of mass attenuation coefficient (μ/ρ), mass energy absorption coefficient (μ_en/ρ), mass energy transfer coefficient (μ_tr/ρ), energy response, and pulse height spectrum. Results indicate that within the 20~50 keV energy range, RW-3 and PMMA exhibit significant deviations from WATER in various parameters, indicating poor water equivalence. Conversely, Blue Water and RW-1 demonstrate good agreement with WATER across all parameters, suggesting better water equivalence. In cases where RW-1 is unavailable, Blue Water can be considered as a suitable alternative. This research provides a foundation for the design and establishment of a device for the absolute measurement of absorbed dose in 1 cm depth water by I-125 particle source.

The China Atomic Energy Research Institute (CIAE) has designed a fully automated extrapolation ionization chamber with variable collection area. Compared with the traditional type of extrapolation ionization chamber, it has the advantages of variable collection area, and the adjustment of the depth of the ionization chamber can be accomplished through the control system, and it can cope with the measurement requirements of different sizes of radioactive sources by providing an adjustable and selectable mode of four sizes of collection electrodes, and the fully automated measurement system is beneficial to reduce the measurement error of reference absorbed dose and improve the efficiency and accuracy of measurement results. The fully automated measurement system helps to minimize the measurement error of the reference absorbed dose and improve the efficiency and accuracy of the measurement results. In this paper, the performance parameters such as leakage current, saturation curve, ion collection efficiency, extrapolation curve, and effective collection area of this extrapolation chamber are investigated to evaluate its feasibility in β-ray dosimetry. Through experimental measurements, the leakage current of the ionization chamber is in the range of 4.2 fA~7.7 fA, which is smaller than the specified 12 fA, the ion collection efficiency is greater than 0.9999, the slopes of the extrapolation curves are 0.6417 and 1.3167 for the irradiation of the two radioactive sources, ⁹⁰Sr/⁹⁰Y and ⁸⁵Kr, which are additionally with beam flattening filters 10 cm from the source, respectively, and the effective collection areas of the four collection poles are 28.35132 cm² , respectively. The experiments show that the performance of this new type of extrapolation ionization chamber is qualified, and its feasibility in β-ray dosimetry is verified.

In recent years, significant changes have occurred in the world situation, with major power games rapidly escalating and unprecedented intense geopolitical conflicts. The new distributed, miniaturized unmanned aerial vehicle (UAV) swarm system has played a crucial role in geopolitical conflicts, demonstrating the strong combat capabilities and enormous combat potential of highly intelligent and autonomous unmanned combat equipment in future joint operations. This paper completes the design demonstration of an integrated electro-optical countermeasure system for air-to-ground combat by analyzing the technical characteristics of "low, slow, small" UAVs and the advantages of electro-optical systems in countering them. Through the analysis and demonstration of strike plans and indicators, methods such as multi-station coordination and adaptive saturation interception can meet the destructive requirements for incoming "low, slow, small" swarm targets.

Industrial control systems play a critical role in the operation of industries. It is crucial to ensure the reliability and security of these systems through safety and security-related event detection. As an emerging neural network architecture, Transformer can effectively extract global dependency characteristics. However, existing Transformer-based methods ignore the extraction of local correlation features, which are essential for safety and security-related event detection. To address this issue, in this paper, we introduce a convolutional Transformer model. The proposed model comprises a convolution module and a Transformer module, dedicated to capturing local correlation characteristics and global dependency characteristics, respectively. Comparative experiments are carried out on an industrial control system dataset. Experimental results indicate that the proposed approach achieves superior performance in both three-class and multiclass detection tasks.

Digital twins serve as a connection between the physical and digital worlds, enabling the creation of a comprehensive digital representation of the entire lifecycle of physical materials, including material selection, structural design, processing, manufacturing, and operation and maintenance management. This paper examines the technical framework and implementation process of equipment digital twin modeling and analyzes the standardization requirements for key technical aspects such as basic rules, information acquisition, model accuracy, timeliness, versatility, as well as model verification and management. Through a case study of a workshop manipulator, the research explores the modeling theory and technology of equipment digital twins, providing a theoretical basis, standard support, and technical references for the application of digital twin technology throughout the equipment lifecycle. Finally, the future development trends and challenges of equipment digital twins are discussed.

Safety and security analyses are crucial for developing safety-critical cyber-physical systems as they can identify potential design defects and vulnerabilities. The Architecture Analysis and Design Language (AADL) is a widely used modeling language for model-based systems engineering, which has well-defined semantics that facilitate safety and security analysis. However, most existing AADL-based analysis methods focus on the safety or security aspect, ignoring the impact of security factors on safety and vice versa. This paper presents a safety and security co-analysis approach for AADL models, namely, fault-attack tree analysis. The proposed method incorporates common security factors that threaten the integrity and availability of the system into fault tree analysis, allowing for comprehensive identification of system risks. The applicability and effectiveness of the proposed method are evaluated through two case studies.

The use of large aircrafts for logistics supply is a key development direction in the future, but the existing research on mission planning problems has few problems and considers the actual environmental factors. This paper mainly considers the actual environmental factors such as aircraft loss and delivery window period, and models the task assignment and path planning of large-scale aircraft for material delivery and supply. On the basis of direct optimization with a basic genetic algorithm, an improved optimization idea and corresponding algorithm combining prior knowledge are proposed, which improves the problem of slow convergence caused by too large search space, and improves the solution speed and optimization effect of the model.

In view of the current situation that the phenomenon of power pilferage is common and the difficulty of anti-theft, a BP (Back Propagation) neural network steal detection method optimized by whale algorithm is proposed. This method detects the power theft behavior through the power trend decline index, line missing index, and problem index, and optimizes it with the whale algorithm since BP neural network algorithm to raise its accuracy. The experimental results show that the whale algorithm capable of optimizing the BP neural network well. Among them, the MAE and MSE were reduced by 0.02, and the RMSE error was reduced from 0.215 to 0.16. Therefore, the model can detect the theft behavior well, which provides a certain reference value for the research of anti-theft.

In order to solve the shortcomings of slow, time-consuming and high labor cost of manual object recognition and archiving tasks, a scene image object recognition method based on YOLOv5 is designed. Based on the YOLOv5 model, a virtual adversarial training mechanism is introduced to enhance the processed data to obtain new extended data to solve the problem of a small number of labeled samples in the semi-supervised learning classification model, and improve the robustness and generalization of the network structure. transformation ability. The experimental results show that the proposed YOLOv5s + VAT detection framework can effectively improve the detection accuracy of the network; the network accuracy can reach 97.042% in the experimental environment under our setting. The detection method can meet practical application requirements.

Due to differences in resource endowments and economic development levels among regions in China, it is crucial to fully consider regional differences and select a reasonable allocation method for achieving the establishment of a unified national carbon market in 2016. This article attempts to seek a new allocation method for carbon emissions rights through the ZSGDEA model, with electricity as input and GDP and population as output. Based on the evaluation results of the model, we compared the differences in quota allocation, fairness, and efficiency between the grandfather method and the efficiency method, providing reference for the establishment and effective operation of a unified national carbon trading market.

Self-training is a popular approach for domain adaptation that relies on generating high-quality pseudo labels. However, due to domain shift, pseudo labels inevitably contain noise. Existing methods usually consider target domain entropy regularization single to correct pseudo label noise. Alternatively, some methods consider how to better mix the source and target domains to mitigate unexpected minimization of pseudo label entropy. But in the mixed mode, the pseudo label presents new uncertainty characteristics and still has space for correction. Therefore, this paper proposes a method that further considers uncertainty of target domain predictions and the use of dynamic thresholding to correct the pseudo label in the mixed domain for better knowledge transfer. Specifically, the proposed pseudo-label correction method uses different loss weights for the mixed domain according to the uncertainty, where the weight of the target domain is determined according to both local uncertainty and overall proportion greater than a certain threshold. Furthermore, a dynamic threshold method is used according to different stages of training and different classes. Our approach achieves competitive results on three adaptive semantic segmentation tasks in Cityscapes→ACDC, Cityscapes→FoggyCityscapes +RainCityscapes, Cityscapes→Dark Zurich in normal to adverse scenes.

Electric vehicle users play a crucial role in the interaction between electric vehicles and the power grid. Therefore, it is highly significant to design a discharge subsidy method that aligns with their willingness to accept compensation, thereby enhancing the incentive mechanism for vehicle-to-grid interaction. This study utilizes questionnaire data from electric vehicle users in a specific city and employs the choice experiment method to construct a model for their participation in grid interaction based on six attributes: discharge season, discharge time, discharge period, discharge amount, and discharge frequency. The random parameter Logit model is utilized to analyze the willingness to accept (WTA) of EV users towards different discharge schemes. These findings provide valuable insights for improving the compensation mechanism associated with EV integration into the power grid.

Aiming at the problem that the static replica management strategy in the cloud storage system is unable to adapt to the changing user access requirements over time, which affects the access efficiency and storage resource utilization of the system, the LSTM file heat prediction model is constructed, so as to effectively extract the information in the heat sequence and alleviate the problems of gradient disappearance and gradient explosion in the recurrent neural network to a certain extent. Suitable appropriate activation function and optimizer are selected to reduce the model prediction error, improve the prediction accuracy and convergence rate, and solve the problem of model convergence instability. On this basis, the dynamic replica management model is optimized, the minimum number of replicas is determined according to data availability requirements, and the LSTM model is used to predict the future replica popularity and define the replica popularity threshold in the system. Based on the future popularity of replicas, the number of replicas for files increases or decreases, while ensuring file availability and saving storage costs. The experimental results show that the heat prediction result of the model in this paper is better than that of the gray prediction model, which proves the effectiveness and progressiveness of the model.

Topic extraction over danmaku is an important task due to the prevalence of danmaku text on the video website. Directly applying traditional topic models on danmaku text can’t work well. The underlying cause is that danmaku text is very short, unconventional and lacking explicit meaning. In this paper, an improved topic model that extends BTM is proposed to infer topics from danmaku text. The special steps in our method are that: (1) a pretraining model is trained on danmaku corpus to obtain word and sentence embeddings; (2) danmaku texts are clustered to generate distinct pseudo-danmaku texts; (3) biterms with the same or similar word pairs are removed from biterm set. Experimental results show that our method can improve the diversity among topics and find some special topic words.

This study conducts a rigorous comparative assessment of YOLOv5 and YOLOv8 for the detection of Demodex mites in microscopic examination images, leveraging crucial metrics such as accuracy, precision, recall, and F1-score. The investigation reveals the unequivocal superiority of YOLOv8, not only in quantitative measures but also substantiated by visual evidence, showcasing its applicability for real-time scenarios. YOLOv8 exhibits exceptional accuracy in overall detection and introduces a novel functionality for quantitative assessment of individual mites, providing essential granularity for precise diagnoses and therapeutic planning within dermatological and ophthalmological contexts. Positioned as a substantial advancement in object detection methodologies, YOLOv8 holds promise for significantly improving both accuracy and granularity in Demodex mite detection within microscopic examination images. While acknowledging potential limitations associated with dataset-specific considerations, this research underscores the imperative for further validation across diverse clinical scenarios. Computational considerations for real-time processing prompt future investigations to explore optimization strategies, particularly in resource-constrained environments. These findings position YOLOv8 as a valuable tool for clinicians and researchers engaged in dermatological and ophthalmological studies, offering heightened accuracy and nuanced insights. Ongoing research, encompassing clinical validations and comparative assessments with other state-of-the-art models, is anticipated to contribute to a more exhaustive understanding of YOLOv8’s potential and limitations in real-world applications based on microscopic examination images.

Strengthening old bridges is an important task in bridge engineering, which can improve the bearing capacity and service life of bridges. Corrugated plate material is a new type of structural material with characteristics such as lightweight, high-strength, and corrosion resistance, which is widely used in bridge reinforcement engineering. This paper aimed to study the bearing capacity of corrugated plate materials in old bridge reinforcement based on computer simulation, and further improve the performance of corrugated plate materials in bridge reinforcement through optimization design. This paper analyzes the characteristics of corrugated plates and proposes finite element modeling. The comparison of different plate materials was conducted using simulation experiments and quantitative and qualitative analysis. The experimental data showed that when the load was 450 kPa, and the numerical result of its bottom displacement was 6.3 mm; The experimental result was 5.9 mm, with an error value of 6.35%. The corrugated plate reduced the displacement of the equipment in the test, indicating that it can better withstand pressure.

With the development of cloud computing, the Internet of Things, and other new technologies, people have put forward higher requirements for virtualization product services and management. Based on the SaaS model, this paper analyzes and develops a set of intelligent management systems based on desktop electronic science and technology ecological projects with practical application examples. The system uses the model framework to realize the collaborative control function between devices by studying data interaction and designing a digital virtual ecological product development intelligent system. Users can select the device and resource type of interest, add or download relevant hardware facilities and other information according to their own needs, and also monitor the storage and management resources in the virtual machine in real time. After that, this paper also tests the performance of the system. The test results show that the development efficiency of the digital virtual ecological product development and design intelligent system based on SaaS mode is more than 83%.

As contemporary computational tools provide more avenues for design development, control, and communication, designers now find it easier to tackle design issues. Furthermore, they are not restricted by material or production limitations in the virtual realm, giving them greater creative liberty. However, this digital liberation has resulted in a gap between cutting-edge digital design ideas and their physical implementation at construction sites, especially in China where resource, skill, and time constraints have widened the chasm between the virtual and actual built environments. This study therefore aims to reveal the disparity between modern intricate digital techniques and standard construction practices, highlighting the significance of employing augmented reality (AR) in construction. The study will focus on Microsoft’s HoloLens to determine if AR can assist unskilled labourers in complex manufacturing tasks. It seeks to show how AR can bridge the gap between high-tech design methods and basic construction procedures.

In this paper, the topology and physical mechanism of the driving stick force feeling system are expounded in detail, the driver control is introduced into the system modeling, and the mathematical model of the driver's skin muscle dynamics is established, so as to further establish a more complete and accurate mathematical model by integrating the whole system, and based on the established mathematical model in this paper, the “human-in-closed-loop” steering stick simulation method is proposed. On this basis, the design of the control system is completed. Finally, the model accuracy of the designed control system, the functional characteristics of the system, the zeroing characteristics of the driving stick, the characteristics of the static force sensing system (torque loading loop characteristics) and the characteristics of the integrated system are simulated and analyzed by MATLAB-Simulink software, and the simulation results verify the accuracy of the model and the effectiveness of the design method.

The Multivariate Correlated Response Vector Generalized Linear Model is an important yet structurally complex class of generalized linear models. Commonly used variable selection methods, such as LASSO (Least Absolute Shrinkage and Selection Operator), are ineffective in efficiently identifying the crucial variables for this type of model. In this paper, we propose a novel variable selection method based on Markov chain Monte Carlo (MCMC) methods, specifically the Random Search method, to address this issue. This approach proves to be an effective variable selection method, as demonstrated by simulation analyses indicating its superior efficiency compared to LASSO and exhaustive subset search methods.

The rapid development of deep learning technology has significantly impacted the fields of Natural Language Generation (NLG) and text composition, leading to numerous noteworthy advancements. These advancements encompass innovations in language models, sequence-to-sequence generation, automatic summarization, automated document generation, and content creation. This paper analyzes the natural language generation and text composition processes within the context of deep learning and text generation, focusing on language models, sequence-to-sequence generation, automatic summarization, automated document generation, and content creation. The application of deep learning models in text generation is exemplified by the construction of a Seq2Seq model based on a Bi-LSTM network architecture, employing a greedy algorithm to achieve rapid text generation.

Spodoptera frugiperda (Fall armyworm, FAW) is one of the major agricultural pests in the world. It is of great significance for ensuring crop yield to accurately recognize FAW larval instars. This research proposes a hybrid model of Convolutional Neural Network (CNN) and Vision Transformer (ViT) for the recognition of FAW larval instars. The model uses CNN to extract local features of larval images and ViT to capture global image dependencies, achieving effective recognition of larval instars. To verify the model proposed in this paper, an image dataset of 5,006 images balancing the larval instars which are from the 1st to the 6th of FAWs is established. The proposed model achieved recognition accuracy of 99.60% and 92.11% respectively in the two cases of using and not using pre-training. The results show that the proposed model achieves higher recognition accuracy than AlexNet, VGG16, ResNet50, ResNet101, and ViT-B. Moreover, it does not rely on pre-trained weights and performs well on relatively small datasets. The research in this paper provides a new technique of the recognition for FAW instars and also provides a reference for other agricultural pest image recognition tasks.

To address the problem that a single discriminant index is difficult to comprehensively evaluate the quality of tangelo, this paper explores a tangelo grading method based on deep neural network and fuzzy decision. Firstly, a tangelo grading model based on deep neural network and fuzzy decision was constructed, and its structure and functional characteristics were given; Secondly, based on Standard Normal Variate (SNV) and Principal Component Analysis (PCA), the spectral space construction mechanism of tangelo dry water defect characteristics is given. Then, a deep neural network model of dry water defect grade of tangelo is constructed; Thirdly, the Soluble Solids Content (SSC) feature spectral space extraction model based on Multivariate Scattering Correction (MSC) and PCA was constructed. A deep neural network regression model for SSC content of tangelo was constructed; Finally, based on fuzzy logic, a fusion decision classifier for tangelo grading was constructed. The experimental results show that the proposed method can realize the comprehensive decision of tangelo quality and effectively improve the utilization rate of fruit.

The existing research on digital image forensics at home and abroad provides certain technical support for the authenticity appraisal of digital images. However, most users have not fully examined the difference between the normal processing of digital images and malicious tampering behavior, which not only stifles the normal processing of digital images such as beautification and polishing, but also hinders the development of digital image processing technology and the promotion of corresponding software. This means that it is not only about simply detecting whether the image has been processed but also interpreting and evaluating the detected processing in conjunction with the image content. Based on the user’s needs for normal image processing, and considering the difference in the authenticity of the image content before and after processing as the standard, correctly distinguishing between normal user modification and malicious tampering, is different from existing digital image forensics technology. This article proposes a digital image authenticity evaluation model based on an encoder-decoder network, which utilizes deep learning techniques to extract relevant features of digital images and construct an authenticity evaluation model. The encoder-decoder network is used to comprehensively evaluate whether the semantic information of digital images is true and trustworthy, thus avoiding misjudgment of authenticity. Finally, the effectiveness of the proposed digital image authenticity authentication model was verified through experiments.

With the continuous maturation of data mining technologies, intelligent fault detection in optical networks has become a prominent area of research in optical network fault analysis. However, due to the complexity of domestic optical networks, intelligent fault detection in optical networks still faces significant challenges. This paper investigates an intelligent fault detection method for Optical Transport Network (OTN) based on the BERT neural network. The research begins with the analysis of raw alarm data obtained from collection tools. The feature data was extracted from the alarm information. Subsequently, employing clustering analysis methods, the paper conducts a clustering analysis of OTN network fault alarm data. The parameters such as alarm time, duration, business network, and element network were considered to categorize different fault work order information. Finally, the fault category information was used as labels for OTN network fault alarm data, and the improved BERT neural network was trained. After obtaining a well-trained BERT model, it is deployed in the live network environment to achieve intelligent detection of OTN network faults.

This study introduces a novel and comprehensive diagnostic approach for Dry Eye Disease (DED) by combining a dedicated Ocular Surface Disease Index (OSDI) questionnaire and a measurement system tailored for Chinese citizens with the implementation of the YOLOv8 deep learning model. The research involves the analysis of 52 real-world ophthalmic videos to detect eye blinking conditions, with the model trained to identify abnormal blinking patterns through feature extraction such as blink frequency, duration, and irregularities. Performance metrics, including mean Average Precision (mAP), specificity, recall, f1-score, and Frame Per Second (FPS), are measured on a PC (CPU, Core i5-10500H) with an input size of 640*640. The integration of these deep learning methods, utilizing both subjective OSDI questionnaires and objective ocular blinking videos, signifies a groundbreaking approach that enhances diagnostic accuracy for DED. The study anticipates transformative effects on DED diagnosis and improved patient outcomes as technology advances. Additionally, the research team introduces a user-friendly system for dry eye detection, named the “AI Dry Eye Analytic System,” accessible at the URL “mini.ac.cn,” demonstrating the practical implementation of the developed methodologies.

This paper analyzes the traditional implementation methods of online analytical processing (OLAP), and on this basis, puts forward the implementation scheme of the subject-oriented database based OLAP (SOLAP), and gives the description of the relevant functional model, the maintenance strategy of the topic library and the implementation method. This scheme can effectively improve the system execution efficiency and reduce the system resource occupation rate by building a topic library and saving the relevant information of user query requests. It greatly improves the working efficiency of traditional OLAP and reduces the occupation of computing resources.

In this paper, we first introduce the fuzzy clustering analysis and application based on AFS (Axiom Fuzzy Set) fuzzy logic, then, we quantitatively analyze the key statistical indicators of the current real estate price and the relationships between the big class indicators and the small by and AFS fuzzy cluster analysis, and open up a new area of cross-analysis between statistical disciplines and fuzzy mathematics. Through AFS clustering analysis, objective classification can be carried out based on the statistical relationships between indicators, and specific classification results can be determined according to the required standards. This allows for further causal relationship fitting and provides a clear path for finding specific functional relationships between indicators, which is helpful for statistical research and analysis, equivalent to principal component analysis in statistics.

In order to emphasize the safe operation of the railroad, it is necessary to carry out periodic inspection of the track and trackside equipment. As the Chidabai Railway intelligent inspection trolley now adopts a long static base alignment time of 5 minutes required during operation, we propose a forward/inverse navigation scheme alignment algorithm, giving detailed steps for the discrete inverse SINS algorithm and discrete Kalman filtering algorithm. Through using 150 seconds for forward and reverse, the alignment time is halved and the heading alignment accuracy is less than 0.3 mils compared with the traditional 300 seconds Kalman fine alignment program, this method can greatly improve the operational efficiency while ensuring the accuracy of data processing.

With the development of the Internet of Things era, the quality of various services in daily life is improving. High-precision positioning capabilities are needed in various industries, such as goods tracking in the factory logistics industry, robot delivery in shopping malls and restaurants, and special population management in hospitals and schools. Due to its large frequency loan and narrowband pulse characteristics, Ultra-wide band (UWB) technology has a strong resolution in both time and space dimensions, so the centimeter-level positioning capability based on UWB is recognized and adopted by various industries. However, in the actual use of UWB, the Non-Line-of-Sight (NLOS) problem is a difficult issue to solve, which greatly affects the positioning experience. This paper proposes an Angle of Arrival (AOA) based UWB-positioning framework using a deep contrastive learning network, referred to as AUCNet. In AUCNet, the Channel Impulse Response (CIR) data of multiple antennas are enhanced and aligned, and then a deep convolutional network and a multi-channel attention mechanism network are used to identify NLOS scenarios. Finally, based on the NLOS identification results and range-angle information, we use a weighted AOA positioning method to solve the final positioning location. We conduct extensive experiments in real sites to verify NLOS identification accuracy, average positioning error, and deployment cost of several positioning methods. At the same time, we conduct an ablation experiment on the deep learning network structure of AUCNet. The experimental results show that AUCNet achieves an average positioning error improvement of 5.1 cm and an NLOS identification accuracy rate improvement of 8.5%.

With the rapid development of e-commerce industry in China, online shopping has gradually become the prevailing consumption mode in recent years. The number of comments on online commodity has also increased rapidly, but the comments are mixed and many of them are often misleading consumers’ shopping decisions, which seriously endangers the environment and market order of e-commerce. In order to extend the scope of the model and reduce the reliance on labeled samples. This paper proposes a transfer model for evaluating text classification based on CNN-RBM. Through the parameter migration of the network layer and the fine-tuning of the model hidden layer, the learned knowledge in the source domain is transferred to achieve classification of the target domain. Due to the negative transfer phenomenon in transfer learning, the performance of the model is reduced. This paper proposes an optimization algorithm based on gradient descent, which is continuously updated in the neural network according to the learning rate, so as to improve the accuracy of text classification on the target domain evaluation. In order to verify the validity of the migration model in this paper, a small number of target domain samples are used to construct the migration model, and the experimental analysis is compared with other models. The experimental results show that the migration model proposed by the text has achieved more accurate classification results on the evaluation text data set.

This paper aimed at the modeling research of the real-time regulation of ambulances in emergency scenarios, an online reinforcement learning algorithm based on the data enhancement method is proposed and applied to the solution of the ambulances control model. The experimental comparison with various algorithms shows that the method proposed in this paper not only has a strong computational advantage in the simulation experiment but also has a strong computational advantage. Moreover, it has practical application potential in emergency rescue operation regulation.

Rice stands as one of the most crucial staple crops worldwide. The traits of rice grains are important indicators for evaluating grain quality. Traditional grain trait measurement methods rely on manual measurement by experts, which not only suffer from low efficiency but also are susceptible to individual subjectivity, resulting in compromised accuracy and stability. To address these issues, this paper proposes an automatic method for measuring grain trait. Firstly, the grain foreground is extracted using superpixel segmentation. Then, a concavity point detection algorithm based on a sliding circular model is employed to detect concavity points on the contours of grain connected regions. Multiple-round ellipse fitting is applied to segment individual grains. Finally, the grain trait parameters are obtained using the minimum bounding rectangle. Experimental trials were conducted on a grain image dataset consisting of three rice varieties: Huanghuazhan, Japonica, and Xiushui 134. Comparative analysis between the proposed algorithm and the SmartGrain software was performed to evaluate their performance. The experimental results demonstrate the robustness and high accuracy of the proposed method.

Aiming at the optimization design of steel cutting, which needs to meet different types of orders and quantity requirements, in order to minimize the number of raw materials used, maximize order fulfillment, reduce the number of tool changes and small machines cutting times, the cutting optimization problem is transformed into the optimal design of the scheme when the raw materials are cut sequentially. First of all, this paper sorts the raw material area and order area, at the same time, the matching dictionary set with the largest rate is established. Secondly, the same type of order is arranged as much as possible, thus reducing the number of knife changes and small machines cutting numbers. Then, the multi-objective optimization model is established. In the simulation stage, through the design of feasible algorithms and programming to achieve an optimal solution, finally, get a good cutting plan, it provides a useful reference for practical application.

Air pollution is one of the serious problems facing China at present. The study of air quality in major cities in China is of great significance for environmental protection and air pollution control. In this paper, the semi-supervised clustering method is used to analyze the air quality in Nanchang City, Jiangxi Province, which provides a new research idea for air pollution and environmental protection in Nanchang City. Firstly, the data are preprocessed and feature extracted. Secondly, the semi-supervised K-means algorithm is used to cluster the data. Finally, the clustering results were evaluated and analyzed by visualization and statistical analysis. Based on the daily average content of atmospheric PM_2.5 in 9 locations in Nanchang in 2021, different types of PM_2.5 were obtained by elbow analysis method, and the characteristics of PM_2.5 pollution in different regions of Nanchang were studied and analyzed by fast K-means clustering technology. The results show that the seasonal variation of PM_2.5 concentration in Nanchang City is very significant, and the overall characteristics are high in winter, low in spring, and low in summer and autumn.

In order to study the temperature distribution and variation law of hot mix asphalt mixture inside the dump truck compartment during transportation, and effectively reduce the influence of temperature segregation of mixed material on pavement construction during transportation, this paper establishes a simulation calculation model of the temperature field of hot mix asphalt mixture during transportation, and analyzes the temperature variation law of the mixture based on nano aerogel insulation felt-slag cotton felt insulation layer from the three aspects of line diagram, surface diagram and body diagram. The results show that: In these three directions, the absolute value of the temperature distribution rate change rate in the low temperature region on both sides of the mixture is large, the absolute value of the temperature distribution rate change rate in the middle position is relatively small, and the temperature change rate of the temperature distribution rate in the highest temperature area is almost 0. At the same time, the highest temperature area of the mixture and the area where the highest temperature is located in the three directions account for a large proportion, and the low temperature area on both sides accounts for a small proportion.

With the rapid development of wind power, it also faces serious safety problems. This paper starts with the brief description, accident analysis, accident causes, prevention and rectification measures of wind turbine fire accident of a wind farm, focuses on the cause analysis of fire accident, and summarizes the causes of this wind turbine fire accident from the detailed analysis of operation data, structural types, test comparison and on-site investigation, and concludes the prevention and rectification measures. Strive to make the wind power industry practitioners have a clear and profound understanding of the accident, and then learn from it.

In recent years, with the rapid development of information technology, the data of electronic medical devices has shown explosive growth. In response to the national call for the step-by-step implementation of the unique identification system for medical devices and the realization of traceability of medical devices. This paper proposes a collaborative tracing method for medical devices based on privacy computing and blockchain technology, to develop data models, architectures, functional modules, and interaction logics suitable for pushing the high-quality development of healthcare security reform, and improving the level of precision and targeting of device regulation, so as to lay the foundation for data sharing. Through the combination of privacy computing and blockchain technology, it realizes full data traceability of medical devices "from production to insured person", secure and trustworthy sharing of data among multiple subjects, safeguarding the privacy and data security of multiple business subjects, and further safeguarding the safety of the use of medical consumables. Strengthen the management of service behavior of medical insurance designated medical institutions, and promote the standardization of medical equipment production and management enterprises.

Due to its complex geological conditions, the water-rich soft rock tunnel will encounter many troubles in the construction process, among which the deformation of surrounding rock has a great impact on the project, and if it is not properly handled, it is likely to lead to serious construction accidents such as tunnel collapse, so it is necessary to analyze the deformation of the rich water rock tunnel. This paper adopts the finite element simulation method and establishes a planar model to simulate the tunnel construction and the deformation process of surrounding rock. The orthogonal experimental design with three factors and three levels is used to simulate nine different working conditions, and the deformation values of vault settlement and arch waist convergence were selected as the observation indexes of the experimental results. The results of the study show that the lower the shear strength of the surrounding rock, the higher the permeability coefficient and the higher its deformation after the excavation method is determined.

This paper analyzes the current research status of the human body 3D reconstruction field, masters its hot topics and evolutionary trends, and provides reference and inspiration for the subsequent application of human body 3D reconstruction in traditional Chinese medicine whole-body examination. The literatures related to human body 3D reconstruction from 2002 to 2022 were selected from the database of China National Knowledge Network. The LDA (Latent Dirichlet Allocation) topic model was used to carry out topic mining, and the research status of human body 3D reconstruction was analyzed from three aspects: recognition and analysis of hot topics, evolution trend of topic intensity and evolution path of topic keywords. A total of 947 Chinese-language literature were included, with research topics identified as digital visual human body 3D reconstruction, human skull and brain 3D reconstruction, human neck 3D reconstruction, artificial bone 3D reconstruction, human heart 3D reconstruction, human hip bone 3D reconstruction, and human liver 3D reconstruction. The hot topics were digital visual human body 3D reconstruction and human liver 3D reconstruction. The topic strengths and evolutionary trends were visualized. The research on human body 3D reconstruction is mainly divided into two aspects: global and local reconstruction, with close connections and integration. The research topic of human neck 3D reconstruction is expected to become a hot topic in the future.

With the rapid growth of machine learning, Machine-Learning-as-a-Service (MLaaS) clusters appear to meet the needs of researchers. However, such clusters generally suffer from high resource utilization and energy consumption. For this reason, most work has focused on resource efficient scheduling for resource consolidation. However, in the process of resource consolidation, job performance is also affected, and it is a serious challenge to minimize energy consumption while ensuring quality of service (QoS) of jobs. In this paper, we analyze the differences in quality of service between offline and online jobs. For the performance interference problem caused by shared resource contention in online tasks, we portray the relationship between shared resources and job performance. As a result, we propose an online learning scheduling algorithm for colocation MLaaS cluster (OLSC), which based on alternating multiplier gradient descent, aims to guarantee the QoS of jobs and minimize cluster energy consumption. The results show that the cluster energy consumption under OLSC is only 59% of RI-FFD and 83% of RS-FFD under colocation MLaaS clusters.

The two main tasks in aspect-based sentiment analysis are aspect term sentiment analysis and aspect category sentiment analysis. In this study, we adopt a unified learning approach to train these two tasks simultaneously, leading to an enhancement in their overall performance. We also introduce three BERT-based models to extract various aspects of semantic information from the context. The best results are achieved by combining these models using a stacking strategy. The experimental results on the MAMS dataset indicate that our model achieve an F1 score of 82.25%, surpassing the performance of neural baseline models.

The current network vulnerability detection model adopts the scanning mode, ignoring the semantic relationship between network nodes, resulting in serious problems of missed and false detection, and low detection accuracy of the model. Model for optimizing the above defects, the Figure convolution neural network based network security holes static detection model. Abstract syntax tree tool is used to build the code attribute graph of network security vulnerability, and the vulnerability is statically analyzed according to the code attribute graph. The vulnerability detection model is constructed by taking the matrix composed of eigenvectors of neighboring basic block and the adjacency matrix of control flow graph as the input of graph convolution network model. In the detection model experiment, the average effective detection rate of the constructed model for vulnerabilities is 96.26%, and the missed detection rate and false detection rate are significantly lower than other methods, and the detection accuracy of the model is better.

In the face of increasing Android malware threats, we conducted in-depth research aimed at addressing classification and interpretability issues. We use dynamic monitoring technique to obtain the information of parameters passed between API call sequence and function. The API sequence is formed into the graph, the API is used as the node, and the parameter information is embedded into the edges. The Graph Attention Network (GAT) is used to manipulate the graph. The attention score generated by GAT also plays a guiding role in node selection in subsequent reinforcement learning tasks, which is used to guide the generation process of simulated malware behavior. In terms of interpretability, we emphasize the importance of explaining GNN model decisions. We propose an innovative approach, the Malicious behavior generator. MB-generator models the interpretation task as a continuous decision process, using a policy network to predict the connection choices of nodes within the function call graph to form a behavior subgraph. The reward mechanism takes into account the dependencies between the newly added node and the previously added node to provide a better explanation of the behavior set. In summary, our research work will have a profound impact on the field of malware analysis and defense. We achieved higher performance on the dataset with 98.43% accuracy, higher than Mamadroid and GDroid. The results show that the proposed method is effective for interpretation graph neural network prediction.

The promotion of electric vehicles can reduce greenhouse gas emissions. However, drivers prefer using traditional cars, because the electric cars have some problems that currently cannot be solved. It cannot fulfill a long-distance driving performance and has mileage anxiety problem. This problem can be eased by installing more charging stations in appropriate locations. In order to determine the location of the public charging station and minimize the total cost while satisfying a certain degree of coverage, this article proposes two optimization models: the complementary partial (CP) coverage method and geometric segmentations (GS) strategy. Two mathematical models are employed to identify the most cost-effective sites for Electric Vehicle (EV) charging stations. This project utilizes the CP method and the GS strategy. Through case studies, it is demonstrated that GS proves to be a more precise approach compared to CP.

This paper proposes a knowledge architecture model to improve the universality and adaptability of the line-seru conversion model by introducing ontology-related theories and methods. A general ontology model is proposed for the line-seru conversion domain, which combines scene and rule ontology. Then, general scenarios and rules are constructed based on their characteristics, and empirical research is carried out with line-seru conversion under multiple scenarios. Based on empirical evidence, this study creates a general reference model for line-seru conversion modeling that can be used in various scenarios. The basic reference model can be easily modified to fit different line-seru conversion scenarios, making it easier to study scenario modeling and production operations.

An advanced multi-strategy hybrid algorithm, denoted as the Butterfly-Inspired Hybrid Sparrow Search Algorithm (BFHSSA), has been developed to address issues related to slow convergence, local optimization, and inadequate precision within the iterative framework of the conventional sparrow search algorithm. The algorithm initiates the positions of individual sparrows using circular chaotic mapping, thereby augmenting the initial diversity and ergodicity of the population. Leveraging the perceptive motion mechanism of the Butterfly Optimization Algorithm (BOA), adjustments are made to the position update formula of the discoverer, resulting in an enhanced global optimization capability for the algorithm. Additionally, a sinusoidal search strategy is introduced to dynamically regulate the convergence speed of individual followers, thereby augmenting the algorithm’s capability to escape local optima. To assess the efficacy of the proposed algorithm, six fundamental test functions are chosen for rigorous simulation experiments. Comparative evaluations are conducted against prominent optimization algorithms, including the Particle Swarm Optimization Algorithm, Gray Wolf Algorithm, Whale Optimization Algorithm, and the original Sparrow Search Algorithm. The experimental outcomes unequivocally demonstrate that the BFHSSA exhibits superior convergence speed, heightened precision, and significantly improved optimization performance.

In the process of detection of rail surface defect with machine vision, crucial step is to denoise and extract feature. Over time, lots of technicist have studied the denoising algorithm and feature recognition algorithm. Nevertheless, there is still some room for studying the accuracy of algorithms. In this paper, an improved algorithm is presented, which applies a bilateral filtering method to denoise ensure that the defect boundary is not weakened, applies grayscale histogram analysis to optimize calculation of the threshold, and applies invariant moment feature extracting to make it possible to accurately identify falling block and collapse defects. The result shows that the detection algorithm of rail surface defects is optimized effectively and the accuracy of recognition is improved.

While encryption ensures the confidentiality and integrity of user data, more and more attackers try to hide attack behaviors through encryption, which brings new challenges to malicious traffic identification. How to effectively classify encrypted traffic without decrypting traffic and protecting user privacy has become an urgent problem to be solved. This paper proposes a classification method for encrypted traffic based on contrastive learning with spatial and temporal feature fusion. Unlike traditional contrastive learning with a single encoder, this method separately designs a spatial feature encoder and a temporal feature encoder to extract distinct features and implements contrastive learning in the same encoding layer through a fully connected layer to obtain a unique feature representation of encrypted traffic. This method focuses on the consistent features from different perspectives within the same sample. The effectiveness of this method is demonstrated through ablation experiments and comparisons with existing works.

Indoor positioning technology is one of the core technologies for location services and artificial intelligence applications and will play a very important role in people’s future lives. In order to solve the location problem in a complex indoor environment, this paper presents an indoor positioning method based on Pedestrian Dead Reckoning (PDR) and map matching. This method first uses PDR method to estimate pedestrian’s walking trajectory. Secondly, PDR estimation errors will be corrected by using the map matching algorithms. Finally, the Kalman filter will fuse the fixed PDR results with the Wi-Fi and geomagnetic fingerprints. Preliminary research shows that this method can provide confidence in the indoor positioning scenes, which can be used in malls, museums, supermarkets, airports, and other places.

In the large span crossing of the Yangtze River Fengcheng-Meili 500 kV transmission line project Transmission Line, Jiangsu province, traction rope had been used as the main element in a one-time continuous deployment scheme for line erection. The traction rope is a steel wire rope with a diameter of 32 mm, a length of 4800 m, and a weight of 20.3 t. The deployment time should be less than 40 minutes. In order to solve the engineering and technical problems, it is necessary to establish a unified analysis and calculation method for the force between the traction rope and the drum. This paper analyzes the forming process of steel wire rope on the drum, establishes a model and explains it, and puts forward a calculation method for the pressure of steel wire rope on the drum during multi-layer winding, which can obtain the radial force at different winding angles and positions, thus laying a theoretical foundation for designing the automatic deployment system and guiding the construction.

Electroencephalography (EEG) signals have significant implications in the fields of clinical medicine, brain research, and neurological disease research. During the collection process, the EEG signals are often interfered with by various factors, such as muscle movement and ambient noise, which cause the decline of signal quality and have a significant impact on subsequent EEG analysis. In this study, firstly a dataset with 50380 EEG signals has been built by mixing clean EEG signals and electromyographic (EMG) artifact signals. Furthermore, an improved DEMUCS method, which is a framework for a speech enhancement model, is incorporated into the process of suppressing EEG signals. Additionally, the encoder and decoder layers, as well as the size of the convolutional kernel, are optimized. The experimental results indicate that the improved DEMUCS model demonstrates excellent performance in artifact removal. Compared with other methods, the algorithm proposed in this article achieved MSE (Mean Square Error) of 0.0699 on test data, showing a significant improvement. This study provides new solutions for the field of EEG signal processing, with the potential to improve the quality of EEG signal data and promote neuroscience and medical research.

This study introduces complex variable element-free Galerkin (CVEFG) method to analyze Helmholtz equation with variable wave numbers. Approximation functions are established by applying complex variable moving least-squares (CVMLS) approximation, thus Lagrange multiplier method is chosen for the enforcement of Dirichlet boundary condition. As a result, ultimate solution equation for CVEFG method applied to 2D Helmholtz equation is obtained through the utilization of the corresponding Galerkin weak form. The astringency of CVEFG method is examined in numerical examples through an analysis of the impact of nodes on relative errors. And CVEFG method has high accuracy and speed.

With the rapid development of internet technology, video service providers have gradually amassed substantial user bases, which in turn generate significant traffic and revenue. However, this also escalates network load, inadvertently leading to a degradation in user experience. By predicting and pushing highly popular videos to users, it is possible to enhance user experience, increase revenue, and alleviate network load. In this paper, leveraging the Pycharm platform and utilizing the Python language in conjunction with a MySQL database, we have designed a machine learning-based video popularity prediction system. This system accomplishes the prediction of video popularity.

Speech Quality Assessment (SQA) is the process of evaluating and measuring the perceptual quality of speech signals. In recent years, the common challenge faced in deep learning is when there is insufficient labeled data for training effective models. Deep learning algorithms, especially deep neural networks, often require large amounts of labeled training data to generalize well to new, unseen examples. In this paper, a non-intrusive method based on meta-learning (Meta-SQA) is proposed for the few-shot learning of SQA. The speech samples which are corrupted by different channel distortions and environmental noises are first converted to the spectrum features then the training and testing tasks are formed for Meta- SQA. The U-Net with accurate localization of microscopic speech of neuronal structures is utilized as a meta-learner. The Meta-SQA obtains prior knowledge with the multiple SQA tasks of known channel distortions and achieves good adaption to the new channel distortions and environmental noises. In terms of person correlation and standard deviation of error measurements, this work achieves better performance than the compared methods.

In response to the inadequacy of timing control methods in adapting to the random changes in traffic flow at intersections and the neglect of the impact of intelligent connected new energy vehicles (ICNEVs), we investigate the effectiveness of a full induction control logic for intersections considering ICNEVs. Firstly, vehicle headway and queue detectors are deployed at each approach of the intersection to monitor real-time traffic conditions. Subsequently, a full induction control logic and corresponding parameters are established based on the clearance and queue status of vehicles at the intersection. The study then considers the influence of different permeability of ICNEVs on the full induction control logic. Finally, a comparison is made among the current state scheme, timed optimization, and full induction control schemes using VISSIM traffic simulation software. Additionally, the study further analyzes the effects of intersection full induction control under different permeability of ICNEVs, thereby validating the effectiveness of the research.

Aiming to assist product development and consumer car selection, this article establishes an evaluation system based on objective evaluation methods in the field of vehicle intelligent cockpit testing and evaluation by using the Delphi method and Analytic Hierarchy Process. Ten mainstream models for sale are selected, and experimental verification and data analysis are carried out based on the established intelligent cockpit evaluation system. The results show that the evaluation system can objectively quantify the level of smart cockpit products and quickly identify product strengths and weaknesses, helping to improve and upgrade smart cockpit products.

Given the low detection accuracy, slow speed, and impoverished comprehensiveness of existing Printed Circuit Board (PCB) surface defect detection in small target detection, a lightweight YOLOv7 improved algorithm is proposed. First, the ordinary convolution operation is replaced by Ghost, the model parameters are significantly reduced, the accuracy is almost no loss, and then the Convolutional Block Attention Module (CBAM) is added. Finally, the Contextual Transformer module is added to the detection header to increase the model’s focus on defect features, which can combine contextual feature information to improve the representation of small target features. Ablation experiments were performed on the PCB datasets provided by the Intelligent Robot Open Laboratory of Peking University, combined with the datasets of 10,000 different PCB defect images collected by ourselves. The experimental results showed that the enhanced model reduced the number of parameters by 54% compared with the previous model, and the detection precision mAP was 97.02%. The running speed is 53.4 FPS, which suits the need for high detection precision and speed in industrial.

Semantic image segmentation has been greatly improved since fully convolution network (FCN) is proposed. Recent work has further optimized the results of semantic segmentation, especially DeepLab, RefineNet and PSPNet. In this paper, we exploit Online Hard Instance Mining (OHIM) algorithm that enables the networks to combine category and instance information for semantic segmentation. Online and automatic mining of the hard instance can make training procedure purposive and efficient. Implementation of the proposed algorithm is simple yet effective, and can be combined with any previous semantic segmentation networks. It is worth mentioning that OHIM does not increase the computation cost in testing phase. We evaluate the proposed algorithm on PASCAL VOC 2012 and Cityscapes dataset, yielding a non-negligible improvement for semantic segmentation.

In the history of aircraft development, the real-time attitude information of an aircraft was reflected through the instrument panel in the cockpit for the first and second generation aircraft. However, since the third generation, the application of Head-Up Displays (HUDs) has centralized these scattered instrument data to show them collectively, achieving a more intuitive effect. This paper focuses on the research of the real-time aircraft attitude display system for a certain type of aircraft HUD simulation process, studies the display algorithm of the system’s attitude indicator, and provides the specific implementation process. Important flight attitude parameters such as sideslip angle, pitch angle, etc., are analyzed and displayed. The motion trend of the attitude indicator allows pilots to understand the flight status of the aircraft itself very intuitively. Therefore, the simulation display of the attitude indicator becomes key in the real-time attitude display system. Especially for the attitude lines, the graphical styles of the zero, positive, and negative scale lines are different, which makes the movement of the attitude scale band relatively complex in the algorithm implementation process. This paper analyzes the overall process and data requirements of the attitude indicator display and presents multiple sets of different basic models in the experiments. At the same time, the paper optimizes the algorithm to improve its efficiency. The experiments show that the algorithm proposed in this paper effectively implements the simulation of the attitude scale band.

In the field of deep learning, object detection algorithms, especially the RetinaNet model based on convolutional neural networks, have garnered considerable attention due to their excellent multi-scale object detection performance. However, improvements in network performance often come at the cost of sacrificing inference speed, particularly when making enhancements to the network structure. This paper focuses on mitigating this negative impact by optimizing the improved RetinaNet model using channel pruning techniques. Initially, we conducted a quantitative analysis to assess the importance of each channel within the network to determine a pruning strategy. Subsequently, the network was finely tuned based on the set pruning rate, effectively reducing the model’s parameter count and computational complexity. It is noteworthy that the pruning process may lead to a decrease in detection performance. To address this, we introduced a novel model recovery method based on channel importance to compensate for the loss in performance. Experimental results indicate that this method significantly improves the model’s inference speed while maintaining a relatively high detection accuracy. The achievements of this research not only provide a new enhancement path for the practicality of the RetinaNet model but also bring a fresh perspective to the field of deep learning model optimization.

For the typical interference signal identification problem, extracting the eigenvalues of the interference signal is a key step, and the number of eigenvalues has an important impact on the accuracy of classification and identification. Ensuring that the eigenvalues have good separation and stability under the condition of fewer eigenvalues is a difficult problem in the extraction of eigenvalues. Aiming at this problem, this paper proposes a dimensionality reduction and extraction algorithm of interference signal eigenvalues based on AutoEncoder. The algorithm uses AutoEncoder to reduce the dimension of multi-dimensional eigenvalues, and uses SVM to train and test the dimensionality-reduced data, and the experimental simulation comparison between PCA algorithm and AutoEncoder algorithm is carried out. The recognition rate of the dimension-reduced eigenvalues of the AutoEncoder algorithm is closer to the recognition accuracy of the unreduced multi-dimensional eigenvalues, however, compared with the PCA dimensionality reduction algorithm, the AutoEncoder dimensionality reduction algorithm has a higher accuracy and has obvious advantages.

A digital image tampering detection method based on a combination of unified local binary mode features and energy deviation features is proposed for digital image tampering achieved using content aware scaling technology. This method describes the changes in pixel correlation within local neighborhoods caused by image tampering by unifying local binary mode features, and achieves the same detection effect as traditional local binary modes using fewer dimensional features. Considering the changes in energy deviation of the image before and after tampering, a unified local binary pattern feature and energy deviation feature are combined to train the classifier using joint features, thereby achieving more accurate and efficient detection results. The experimental results show that this method can effectively detect content aware tampering in images and accurately locate tampered areas in Seam Insertion.

This paper investigates how the sea surface temperature (SST) and salinity (SSS) change with the seasons based on K-means clustering analysis. For index k, the elbow method is used to estimate the value. Then this paper analyzes the clustering of SSS and SST in a specific region that contains both characteristic SST and SSS features and compare the clustering figures from each season. Meanwhile, by comparing K-means cluster results computed from raw data to those obtained for classified histograms, we find that it is ideal to combine these two types of data to do an analysis. To illustrate our methodology, we use the Silhouette Coefficient to find the goodness of our K-means cluster, and the calculated Silhouette Coefficient equals 0.8, which illustrates our model has relatively high quality and accuracy. Lastly, we apply other clusters, such as DBSCAN and Gaussian Mixture Model to the same data, but the results indicate the difference between our data is too weak to be clustered by these two methods.

A multi-path traffic scheduling algorithm is proposed to solve the problems of frequent elephant flow collisions, serious link congestion and load imbalance in data centers. This paper makes full use of the advantages of multi-path network topology and software defined global view of the network. First, the sFlow controller is used to detect the size flow. Then, the paper uses the improved k-short path algorithm to increase the non-shortest path, calculates the criticality of each path, and selects the path with smaller criticality as the candidate path, so as to reduce the scheduling algorithm overhead. Finally, the optimal path number and distribution weight are calculated continuously, and the link transmission cost and pheromone in ant colony algorithm are updated according to the calculation results, so as to complete the multi-path traffic scheduling. Simulation results show that the proposed algorithm can improve the network bisection bandwidth and reduce the maximum link utilization and average transmission delay in different traffic modes.

Blockchain technology is a distributed ledger technology that can be employed to safeguard user privacy and ensure the integrity of data, thereby enhancing information security. Traditional algorithms, such as Raft, have been criticized for their limited ability to provide sufficient Byzantine Fault Tolerance (BFT) and vote splitting in this context. To overcome these issues, PS-Raft relies on three key innovations. Firstly, it limits the number of potential Candidates by carefully selecting nodes based on their communication conditions. Secondly, it utilizes an SM2-based proxy signature algorithm to verify the legitimacy of Candidate and Leader identities. Lastly, during term replacement and log replication, it ensures the honesty of Leader by verifying log data. We conducted a local cluster simulation that compared the performance of PS-Raft and Raft consensus algorithms. The results of experiment demonstrate that PS-Raft effectively screens and defends against Byzantine nodes while maintaining comparable overall performance to Raft. This improves hardware and time efficiency for Byzantine nodes attempting to implement such attacks. Furthermore, the PS-Raft algorithm significantly reduces the election time and resolves the problem of vote splitting.

To address the issue of poor performance of anomaly detection models when the number of labelled logs is small, this paper proposes a semi supervised log anomaly detection method LogALST based on active learning and self-training. Self-training predicts a large number of unlabelled logs to generate high confidence log data, reducing the manual labelling cost of active learning. Active learning uses a sample sampling strategy to select log samples with high uncertainty, reducing labelling errors in unlabelled logs during the self-training prediction stage. Through a series of experimental verification, LogALST outperforms existing semi supervised log anomaly detection methods. At the same time, LogALST is also superior to log anomaly detection methods that rely solely on active learning or self-training, achieving better anomaly detection performance with less manual labelling costs.

As the prediction structure of dam slope deformation is mostly unidirectional and the prediction target is not clear, which is easy to lead to excessive prediction error. BP autoregressive algorithm was used to design the dam slope deformation prediction method. Firstly, the prediction data set was established based on the actual prediction requirements and standards. Then, the specific prediction targets were set using the multi-level prediction form, and the multi-objective three-dimensional prediction structure was established. Finally, the deformation prediction model was built through the BP auto-regression algorithm, and the more accurate prediction was achieved by using the dynamic regression correction. The experimental results show that the proposed dam slope deformation prediction method based on BP autoregressive algorithm has higher prediction accuracy and stability, compared with the traditional BP neural network, Beidou GNSS and multi-cluster echo slope deformation prediction methods, the RMSE, MAPE, and MAE of predicted values are only 0.35mm, 5.78%, and 0.32mm, respectively, which can provide high-precision data for dam safety monitoring.

Ecosystem theory is a methodology that uses the interrelationships between ecological elements to improve the efficiency of the entire ecosystem by integrating ecological factor resources. Ecosystem theory applies to all disciplines in various fields and helps to solve the problem of factor separation and inefficiency. Similarly, the theory is also applicable to applications such as on-demand distribution of mine gas flows. On-demand distribution of airflow is the basis of intelligent ventilation and plays an important role in ensuring safety. Existing air distribution algorithms have the problem of artificially selecting adjustment branches and individual adjustment forms. Applying the theory of ecosystem, based on the critical path theory and the cut-set theory, this paper proposes the node pressure energy method to distribute the airflow on demand and analyzes its feasibility.

In response to the challenges of low accuracy, difficult feature extraction, and poor real-time performance in traditional gearbox fault diagnosis algorithms, this paper proposes a gearbox fault diagnosis algorithm based on convolutional neural network (CNN) and fused coordinate attention mechanism (FCAM) stay, a dataset of various gearbox fault images is collected and divided into training, validation, and test sets. Then, a lightweight MobileNetV2-based neural network model with the FCAM is constructed using Pytorch Framework. Subsequently, the training and validation data are input into the CNN with the fused attention mechanism for fault diagnosis model training, and the trained recognition model is saved. Finally, this model is used for the diagnosis and recognition of gearbox fault images in the test set. This algorithm is capable of automatically extracting gearbox fault features and classification, offering advantages in terms of reliability and real-time performance.

The Heisenberg uncertainty principle indicates the position and momentum for one particle cannot be determined at the same time. In signal processing, it points out that the frequency domain resolution and time domain resolution of one signal cannot be too small at the same time. As a general form of Fourier transform, the linear canonical transform (LCT) has made great contributions in signal processing. In this paper, an uncertainty principle for n-dimensional LCT, based on which some generalizations for LCT are obtained.

In times of pandemic, the face-to-face and hybrid Modalities implemented in different Philippines Higher Education Institutions (HEI) were forced to shift to full online. In the 2T of 2022-2023, CHED (Commission on Higher Education) mandated the Higher Education Institution (HEI) to adopt full face-to-face classes or hybrid learning in reference to CHED CMO no 16. Mapúa Malayan Colleges Laguna (MMCL) is a Higher Education Institution (HEI) that adopted “Mapúa MCL FLUX (MMCL FLUX). The research aims to answer the question, Is there a significant difference between the Student Academic Performance in Consideration of Mapúa MCL Flux Modalities (Face to face Modality, Hybrid Modality A, Hybrid Modality B, and Hybrid Modality C) for Dynamics? Researcher finds out that there is no significant difference in the Students’ Performance Under Different Modalities of Mapúa MCL Flux for Dynamics Course at 5% level of significance. Hence, students can choose the most favorable modalities according to their preferences.

X-ray fluorescence data are not only an important indicator in petroleum exploration, but they are also an advantage in lithology discrimination. From the perspective of the application of nuclear technology in drilling and exploration, in this paper, the X-ray fluorescence count rate of the core is used as the basic data for core logging, and the geology and multivariate statistical data are analyzed to analyze the lithology of the core. X-ray fluorescence analysis can provide the fluorescence count rates of 35 elements in a core sample. If there are n core samples, an n*35 matrix can be created. If 35 types of fluorescence count rates are used as the independent variables to directly identify the lithology, the dataset is very large, and some of the independent variables are not strongly correlated. Therefore, the discrimination effect is insufficient for achieving the purpose. Therefore, based on the diagenetic characteristics of sedimentary rocks and the main rock-forming elements and using multivariate statistical methods, in this paper, the count rates of five elements (i.e., Si, Fe, Al, K, and Ca) are selected as independent variables for the lithology discrimination. It is reasonably transformed to obtain the matrix coefficients before dimensionality reduction. The original five-dimensional dataset is reduced to two dimensions using matrix coefficients. Then, Fisher's discriminant method is applied to the dimensionally reduced data to discriminate and classify the lithology. Based on the discrimination results, the X-ray fluorescence count rate and mathematical statistics can be used to obtain a good effect in lithology discrimination

The accuracy of foundation settlement monitoring is affected by many factors. To reduce the influence of the change of datum points in the later period, this paper proposes a multi-sensor data fusion technology based on static level and GNSS for ground settlement monitoring and designs several datum points to find out the more stable datum points through the limit difference test method based on the average gap method. Furthermore, the elevation correction of settlement monitoring points monitored by static level is conducted based on the collected elevation changes of datum points. Additionally, methods to reduce errors according to the source of errors are proposed. This technology is applicable to complex areas such as mountainous areas and tunnels.

[Methods]The binary tree was used to Preorder traversal the relative pressure difference of each node in the irrigation community pipeline network, solving the optimal distance problem between two points with real weights.[Results] Through the analysis of practical application case, using the binary tree first sequence traversal method respectively for different specifications of branch pipe, wool, water flow solution irrigation village pipe network system pressure difference, the results and the calculation results in other literature, the optimal scheme is consistent with the actual situation, shows that the binary tree traversal method in solving the optimal distance of real weight is feasible.[Conclusion] The binary tree traversal method has the advantages of simple principle, convenient storage and clear structure, which can be widely applied to the problem of optimal tree structure with real weight.

Using self-coupling PID (SC-PID) control technology, a pressure control problem for a natural gas supply system was studied. Firstly, based on the natural gas supply system simulation model built under multidisciplinary unified modeling language Modelica, the impact of pumps and compressors on the pressure of the supply system is analyzed through the simulation of the key work flow of the natural gas supply system. Then, a buffer tank pressure controller was designed by self-coupling PID control technology, and the effectiveness of the controller was verified based on the natural gas supply system simulation model. The simulation results verified that designed pressure controller can make the safe and stable operation of the whole natural gas supply system under operating conditions such as fuel supply mode switching and load changes of the main engine.

When applying reinforcement learning to multi-agent decision making, it has been challenging to learn effective strategies in dynamic complex environments. Agents need to consider and interpret the high-level information abstracted from different perspectives of the scenario, which will help them take appropriate actions. However, many existing methods are devoted to learning the end-to-end “observation-action” mapping. In fact, the real-world scenarios usually involve many interacting agents, which makes appropriate end-to-end mapping difficult to obtain. Therefore, this paper proposes a mechanism of multiple scenario-semantic exploration (M2SE) for multi-agent cooperative learning. Based on this mechanism, a multiple scenario-semantic exploration network is proposed to acquire the high-level scenario-semantic information embedded in agents’ observations and to provide agents with heuristics from different views. The M2SE network constructs a “observation-semantic-action” mapping and enables agents to learn more appropriate strategies by fusing the high-level information in the semantic space. It is straightforward to integrate the M2SE into existing MARL algorithms. Experimental findings indicate a notable performance improvement in MARL algorithms when using the M2SE.

Due to the fact that the Hydraulic Hybrid Vehicle (HHV) which has a double-function, energy management and stability control, is a great advantage in the HHV, it is necessary to research the coordinated control between regenerative braking and stability. The research of the coordinated control between regenerative braking and stability of HHV is at the forefront of the study in this aspect. Based on the above factors, in this study, we propose an algorithm of regenerative braking and stability coordinated control. Firstly, we put forward the assembly institutions of the Parallel Hydraulic Hybrid Vehicle (PHHV), then we adopt the logic threshold algorithm to realize the distribution control and use the incremental joint PI control with variable parameters to realize the vehicle electronic stability control system (ESC) control. Simulink/Simscape in the MATLAB software is utilized for the offline simulation of the control algorithm of PHHV. The result of the simulation shows the effectiveness of the regenerative braking force distribution and ESC algorithm.

Traditional spices quality testing methods face problems such as long testing latency, high testing cost, complicated testing process, environment-dependent inconsistency and great influence of human factors on the testing results. In order to solve these problems, this paper proposes a hybrid model based on Shuffled Frog Leaping Algorithm (Shuffled Frog Leaping Algorithm, SFLA) and Back Propagation (Back Propagation, BP) neural network, namely SFLA-BP. The model first obtains the weights and biases as initial parameters through the BP neural network, then uses the SFLA to optimize the parameters, and finally feeds the optimized parameters back into the BP neural network for model training. Experiments show that the hybrid neural network model can effectively improve the accuracy of identifying fragrances and flavors, and reduce the interference of environmental and subjective factors, which has significant practical application value.

In the era of data-driven services, traditional book services and contents have been unable to meet the dynamic needs of students, which puts forward higher requirements for the book information services provided by smart campus. In order to enrich the library’s collection resources and formulate a scientific and reasonable purchasing plan, Douban’s personalized book recommendation scheme can put forward a new management idea for the construction of intelligent library resources. This study takes the books and user information obtained from Douban as an example, uses the Kmeans clustering algorithm to sketch the user information, and combines the collaborative filtering algorithm to provide personalized book recommendation services for the purchase plan of smart campus books.

Dual-nozzle-flapper pressure servo valve (DNFPSV) is frequently used in the field of aircraft landing gear brake system, which belongs to a typical complex electro-hydraulic servo system. Gradual wear and fatigue will occur in actual service, which will inevitably lead to the deterioration of the key parameters of the servo valve and eventually lead to the deterioration of property to the valve. Taking the DNFPSV as the research object, this paper first explains its working mechanism, and completes the whole valve mechanism model by using the lumped parameter method; Secondly, the delivery pressure of the servo valve is closed loop fed back to the input, and an integer order PID controller and a fractional order PID (FOPID) controller based on particle swarm optimization (PSO) are designed for the closed loop control of the DNFPSV; Finally, the FOPID simulation model of DNFPSV based on PSO algorithm is built and the simulation analysis is completed. The simulation results show that compared with the integer order PID (IOPID) algorithm, the FOPID control algorithm based on PSO has better dynamic and static control effect, and can be used for the effective control of DNFPSV.

With the acceleration of the great changes which have not occurred in the world for a century, global uncertainty continues to increase, and profoundly affects the development of smart cities. Facing various uncertainties in the future, the theoretical system and methodology of China’s smart city planning need to be improved and transformed urgently. It is necessary to go beyond the simple and complex paradigms, to gain a deeper understanding of deep uncertainty issues, and to develop new analytical tools. We provide a new framework to the open problem, trying to the smart city adaptive in the deep uncertainty environment.

Regarding the issue of abnormal sales in online shopping websites, the utilization of the MAD-GAN model based on LSTM is considered for handling it, owing to its adeptness in detecting temporal data patterns. The MAD-GAN employs Gan Loss (cross-entropy loss function), which possesses issues of asymmetry and gradient vanishing. To address this, an enhancement using Wasserstein Loss is proposed (referred to as WLoss), resulting in a model termed MAD-WGAN. WLoss asserts that the optimal state is achieved when the data distribution Q aligns perfectly with the sample distribution P. This implies treating the data distribution Q as the true distribution R. However, a certain disparity exists between the actual data distribution Q and the true distribution R. Consequently, a novel definition of optimal distance, termed Wasserstein Deviation Loss, is introduced. Wasserstein Deviation Loss posits that the best state is achieved when a minor difference exists between data distribution Q and sample distribution P (regulated by a parameter β). This state is labeled as WDLoss. To assess the effects of these improvements, a MAD-WDGAN model employing WDLoss is proposed and compared with MAD-GAN and MAD-WGAN. Across the same test dataset, MAD-WDGAN outperforms MAD-GAN and MAD-WGAN models by 5.42% and 0.45%, respectively, in terms of accuracy. This suggests the advantageous nature of WDLoss.

AC-DC coordinated control system is an important technology that can deal with the power exchange and transfer between different power systems. The system controls the operating parameters of AC and DC transmission lines, such as line voltage, phase Angle and power flow, to maintain stable grid operation. The stability control strategy is a very important part in the AC-DC coordinated control system of power grid. It means to adjust the operating state of the power grid by controlling the operating variables of the system to ensure that the power grid can maintain a safe and reliable operating state under both transient and steady state conditions. In this paper, the deep learning theory is applied to the electromagnetic field analysis of the power grid, and the deep learning model is trained and built with the electromagnetic field distribution data corresponding to different power grid structures, so as to predict the electromagnetic field distribution instead of the traditional finite element calculation. Based on the deep learning theory, the convolutional neural network, The in-depth study of the computing process and network structure of the short-duration memory network and its variants determined the hyperparameters needed to build the deep learning model, then established the deep learning model and trained the model through samples, compared the models through evaluation indicators, selected the optimal model, and introduced three different test functions to test the performance of the model. Finally, the optimal model is used to predict the electromagnetic field distribution.

Currently, internet and distance education technologies are rapidly integrating. Learner-centered online learning offers the advantages of flexibility in time and space, affordability, and comprehensive services. However, learners often face the challenge of dealing with an overwhelming amount of learning resources, which is commonly referred to as “information overload.” This paper aims to address the “cold start” problem in the collaborative filtering (CF) algorithm by designing and implementing a personalized learning resource recommendation system. This system incorporates the CF algorithm and introduces the latest learning resource list, popular learning resource list, and user resource tag matching. Additionally, deep learning technology is applied to recommendation systems to tackle issues of data sparsity and information overload. This article studies the characteristics of DeepFM and NeuMF models and embeds them into the model using one hot encoding. DeepFM training results in the model have the lowest Loss function and the highest AUC. In the context of information overload, personalized recommendation services for learning resources in the internet learning environment have become an inevitable trend, with important research significance and wide application value.

Robots have played an increasingly important role in various fields of society such as industrial manufacturing, family life, national defense and security, while also placing higher demands on their accuracy, real-time performance and efficiency. Object detection and grasping pose estimation are the basic abilities that industrial robots should possess, playing a crucial role in industrial sorting, palletizing, assembly and other work. Deep learning technology extracts features through neural network, which can greatly improve the detection speed and accuracy under the guarantee of big data and computing ability. In this paper, through the research of target detection and manipulator grasping algorithm based on convolutional neural network (CNN), a multi-target tracking algorithm combining YOLOv3 target detection algorithm and kernel correlation filter (KCF is proposed. The experimental results show that the KCF algorithm has good tracking performance and good fault tolerance for small amounts of occlusion. Thirdly, this paper proposes a visual guidance method for robot grasping combined with Kalman filter prediction, which can guide the robot to accurately complete the grasping task. Finally, this article built a robotic arm sorting system and ported the YOLOv3 network to the Xavier platform. The experimental results show that the network can effectively detect and recognize targets in the experimental environment. The test results verify the effectiveness and availability of the target sorting system.

Facing the fierce competition in the construction market, the contractor’s investigation and forecast of construction material prices is not only an important prerequisite for reasonable quotation and winning bids, but also a key link to ensure the profitability of the enterprise. The traditional cost management methods are gradually not applicable to the current cost management. With the development of science and technology, the emergence of BIM and BP neural network has brought a new concept to cost management. In this paper, based on BIM technology and BP neural network model for ex ante cost prediction; after the completion of the project, ex post cost deviation analysis is carried out to analyze the predicted cost with a certain engineering project as a case, which verifies the reliability and scientificity of the scheme and has a greater practical value for cost control and social significance.

Industrial structure refers to a series of economic activities that, within a certain space, play a certain role in promoting economic development and can have a certain impact on other industries. It is closely related to scientific and technological innovation, technological progress, etc. In the process of China’s economic development, the issue of industrial structure has always been one of the important factors restricting economic development. The purpose of this paper is to study scientific and technological innovation and industrial structure optimization and upgrading based on the improvement of KNN-DPC algorithm. On this basis, this paper proposes an improved algorithm KNN-DPC algorithm, which uses second-order recursion to maintain approximately linear motion between the input model and the vector generated by the objective function, and then combines the weight matrix solution method to obtain the equation and correct it. Finally, the purpose of improving the prediction effect and convergence speed of the model is achieved. In turn, this problem is solved so that the input result is consistent with the output, that is, the optimal solution state is obtained under the same circumstances. This algorithm is used to model and analyze some urban spatial variables in spatial statistics, combined with the analysis of the country’s own scientific and technological innovation capabilities and industrial structure in recent years, to provide a decision-making basis for industrial technology development.

The power company adopts the agreement inventory bidding procurement method, through the management of agreed procurement material requirements and problems in the implementation of the agreement, it can develop an optimized implementation plan, strengthen the scale of centralized procurement of power materials, improve efficiency, reduce help to establish a reasonable material procurement management standard for power enterprises, and lay a solid foundation for ensuring the supply of materials for distribution networks. The purpose of this paper is to study the agreement inventory procurement strategy based on the decision tree algorithm. According to the actual needs of m company, a material inventory control and auxiliary decision-making system based on decision support technology are designed. Through C4.5 decision tree algorithm, the decision support proposal can reduce the occurrence of material inventory shortages and greatly reduce the complexity of the work. It can effectively improve the level of corporate inventory control, reduce corporate operating costs and improve the core competitiveness of enterprises.

Exoskeleton motion control technology has become an important industrial enhancement technology. However, in practical application, due to the complex human mechanical structure and changeable working environment, the existing exoskeleton motion control platform and algorithm still have some limitations and shortcomings, which need to be further optimized and improved. Based on the three-dimensional model and data analysis of the exoskeleton motion platform, a new motion control algorithm is proposed, verified, and improved. Through data analysis and experimental verification, it is found that the new algorithm can significantly improve the mechanical performance and control performance of the exoskeleton motion platform. Specifically, the average speed, average acceleration, and posture change angle range of the new algorithm are increased by 0.13 m/s, 0.28 m/s, and the time series of task execution is shortened by 639.3 ms. This study provides new insights and optimization methods for the construction of exoskeleton motion control platform and control algorithm, which will further promote the development and practicality of this field.

Automated material transportation systems are an important component of modern intelligent algorithms, and their multiobjective scheduling management plays an important role in improving material transportation efficiency and reducing costs. This article combines a method based on big data and intelligent algorithms to apply to multi-objective scheduling in automated material transportation systems, and combines it with the material transportation system. Firstly, intelligent and robotic technologies are used for data processing and discovery in response to the massive data volume of material system transportation. Then the intelligent algorithm is applied to the material transportation scheduling, which uses genetic algorithm and Tabu search algorithm to achieve multi-objective scheduling. Finally, the application was validated in a practical material scenario system, and the experimental results showed that the sensors achieved 53%, the controllers achieved 49%, and real-time monitoring achieved 36% in material transportation.

The optimization design of shielding for on-board nuclear power supplies is a multi-parameter and multi-objective optimization problem, and the design of key parameters such as shielding layer material and size has an important impact on meeting the requirements of miniaturization, lightweight, and mobility of on-board nuclear power supplies. Using the elite strategy of the Non-Dominated Sorting Genetic Algorithm II (NSGA-II), with the overall objective of optimizing the total mass and dose rate of the shielding layer, the radiation shielding design scheme under operation and transportation conditions is given. The analysis results show that the mass and radial dose of the radial optimization scheme obtained in this paper are 78.08% and 52.47% of the initial scheme, respectively, while the mass and dose values of the axial optimization scheme are 83.52% and 78.39% of the initial scheme, and meet the mass and dose limits. The NSGA-II algorithm is used to adaptively optimize the key parameters such as material and size of the shielding system for on-board nuclear power supplies, and a shielding optimization scheme with light weight and small volume is obtained. This can provide theoretical and technical support for the optimization design of radiation shielding for on-board nuclear power supplies.

With the progress of information technology such as computer application technology and remote sensing technology, real-time meteorological disaster monitoring and early warning and emergency decision-making based on remote sensing data have become possible. There have been a lot of studies on NN (Neural Network) forecast modeling and climate analysis in intelligent early warning, but in meteorological forecast modeling, it is difficult to determine the initial weight and threshold value by using NN method, and the determination of network structure and parameters needs to be trained for many times, which is likely to cause the problem of over-fitting, which greatly affects the generalization ability of the network. In addition, there are local minima in the NN model, which cannot guarantee prediction accuracy and are not conducive to the real-time and effective tracking and prediction of the weather system. Therefore, this paper uses GA (Genetic Algorithm) to optimize the structure of NN and its connection weight and gives the concept of intelligent genetic arithmetic. That is, combining genetic arithmetic and BP arithmetic to optimize the weight of NN, so as to avoid the calculation results into local optimization, and reduce the number of iterative operations, improve the speed of solution. In this article, genetic arithmetic is applied to the meteorological disaster risk warning system, and the NN warning model based on the optimization of genetic arithmetic is established. The experiment result expresses that the average relative error of the intelligent GA early warning model is 0.43%, and the prediction accuracy has obviously improved. This shows that the method is feasible and effective for complex systems. Intelligent genetic arithmetics also have their advantages, and the predictive ability of NN is enhanced.

Aiming at the issues of low contrast, weak edges, and poor clarity of phase contrast cell microscopic images, which lead to the inability to observe cell features clearly, a phase contrast cell microscopic image enhancement algorithm based on sliding window and multi-scale Gaussian filtering is proposed. Firstly, an adaptive sliding window grayscale enhancement method is proposed to enhance the contrast and edge features of phase contrast cell microscopic images. Secondly, in the multi-scale feature fusion stage, the enhanced phase contrast cell microscopic images are filtered with a guided filter to suppress some noise and impurities. And the multi-scale Gaussian filtering weighted fused method is proposed to extract the structure image that contains edge and detail features. Finally, in the structure features enhancement stage, the guided filtered phase contrast cell microscopic images are differential of the structure features constrained by the adaptive enhancement factor to obtain the enhanced phase contrast cell microscopic images. The experimental results show that the proposed phase contrast cell microscopic image enhancement algorithm based on sliding window and multi-scale Gaussian filtering is superior to the existing traditional methods in subjective and objective image quality evaluation, and effectively improves the contrast and enhances the detailed features and edges of the cells in the phase contrast cell microscopic images.

Traditional FCM (fuzzy C-means) image segmentation considers only the image’s pixel value and ignores other factors, which leads to the under-segmentation problem. This paper proposed a color image segmentation method based on the superpixel region features (SRFFCM), which uses the texture, mean, and color features obtained from the superpixel region to construct a new image segmented by the method. Firstly, we use the superpixels method to segment the image into multiple regions. Secondly, to obtain the texture features of the regions, we designed a calculation method of a gray cooccurrence matrix for the irregular regions. Finally, we treated the region as a standard pixel and used the proposed method for clustering under the construction of the irregular regions’ multidimensional vector. Due to the utilization of multiple features of the region, the segmentation accuracy (SA)and mean Intersection-over-Union (mIoU) of the proposed method is more than 90% through the test of images segmentation, and the time complexity is lower than that of most other algorithms.

The high intraclass variations and high interclass similarities make high spatial resolution (HSR) image interpretation very difficult. In this paper, we provide a feature selection method for scene classification, particularly to recognize the images from the industrial scene. The proposed method is based on scene overlapping rate and a quantitative feature assessment metric, considering the interclass distinguishing capabilities of features. Thus, the optimal feature representation for diverse scenes according to the discriminative characteristics of intraclass images can be selected. Specifically, an efficient scene recognition algorithm is presented to identify the easy-confused scenes from the dataset. An evaluation metric is also proposed metric to quantify the distinguishing capability of features with full statistical support. Experiments comparing 14 state-of-art classification methods on two challenging benchmark scene datasets (UCM and AID) show the effectiveness of the proposed method for HSR scene classification.

To meet the needs of farmland surveys and area calculations in agricultural insurance claims, “Jilin-1” satellite video data was used for crop area calculations. Firstly, a set of video super-resolution reconstruction techniques was designed to address the issue of low-resolution video from the satellite. The Walsh-Hadamard transform and deep learning techniques were used to achieve super-resolution reconstruction of the satellite video. Secondly, the reconstructed data was used for crop area calculations. The experimental results showed that the Walsh-Hadamard transform and deep learning techniques can achieve good super-resolution reconstruction of satellite video, and the loss rate in crop area calculation using “Jilin-1” satellite data is within 0.23%, which meets the requirements of area calculation in agricultural insurance.

Face recognition technology is developing rapidly, and the issue of personal privacy protection has become increasingly prominent. How to protect personal privacy while ensuring recognition accuracy has become the focus of research. This study uses deep learning as the basis of face recognition technology and extracts high-dimensional feature representation of face images through the training and learning of deep neural networks. At the same time, for privacy protection, this study proposes anonymization and desensitization technology, privacy protection generation model, distributed learning and federated learning, multi-factor authentication, and other methods. These methods aim to increase information entropy, reduce personal identifiability, and protect personal privacy. Research results show that the face recognition accuracy of this method is 91%-98%. Face recognition technology based on deep learning has made significant progress in improving recognition accuracy and increasing information entropy. This method provides a foundation for efficient and accurate face recognition, and also provides a more reliable means for personal privacy protection.

Recently, driven by advancements in deep learning, there has been substantial progress in object detection technology. Nonetheless, detecting objects in aerial images remains challenging due to the abundance of small, hard-to-detect objects. Although the multi-scale feature fusion strategy of the feature pyramid network (FPN) broadens the detection scale range, partially alleviating the challenge of small object detection, the cross-scale feature coupling constrains further performance enhancement. To overcome these constraints, we propose a novel Detailed Feature Pyramid Network (D-FPN), specifically designed to improve the efficacy of detecting small objects in aerial images. D-FPN comprises two essential components: the high-resolution feature expansion module (HRFE) and detail information enhancement module (DIE). HRFE extends high-resolution features beneficial for detecting small objects via an asymptotic fusion method. DIE supplements detailed small object information by merging high-resolution features with the backbone’s low-level features. We evaluate D-FPN using the VisDrone2019-DET dataset and compare it with current state-of-the-art object detection methods. Experimental findings validate the effectiveness and advancements achieved by D-FPN.

As an important means to assist doctors in the diagnosis and treatment of brain diseases, brain images play a crucial role in the field of medicine, and how to remove the noise in brain images so as to obtain clear images is a problem that many scholars are working on. Traditional filtering methods have the problem that the details of the image are easily damaged in the denoising process, with obvious noise residues, and the corresponding filters need to be designed according to different noise types. In this paper, an improved dual-channel residual convolutional neural network model based on a deep convolutional neural network is cited for removing noise from brain images. The experimental results show that the average peak signal-to-noise ratio of this method is improved by 0.19783 dB to 17.92 dB compared to several other methods, the ability to reconstruct clear brain images while effectively removing various levels of random noise has some theoretical implications in the medical field.

In the modern digital society, mobile phone screenshots are essential for documenting data, proving events, and providing legal evidence. However, modern digital technologies have made taking screenshots easy. Significant differences exist between screenshots and traditional mobile photography images in the formation way, leading to immature industry research in screenshot source tracing, by complicating the authenticity proof. The study aims to explore critical methods and techniques in the source tracing and authentication of mobile phone screenshots. The paper involves collecting screenshot images produced by 50 common mobile phone models available in the market and analyzing their metadata. This research employs a multifaceted analysis of screenshots, using metadata, image characteristics, file naming patterns, and unique markers to trace their origins to specific mobile and working versions. Based on this, the paper further discusses the main challenges faced in mobile screenshot source tracing and authentication. Results show that, despite challenges, integrating digital forensics and multidimensional feature detection enables reliable tracing and authentication of mobile screenshots. The research provides essential technical support and data backing for the authenticity proof of images in forensic science or legal examination and the source tracing of mobile.

With the rise of network cameras, the authenticity of video data they capture has become a common challenge in forensic science and expertise. There are many camera manufacturers and brands available, and videos from different sources have their distinct properties. To evaluate these videos needs further exploration of the key traits. This paper collected eight commonly used network cameras on the market. It studied and analyzed the features of network camera video data from four parts, i.e., storage media traits, video properties, metadata features, and image details. This research builds effective authenticity-proof techniques for webcam recordings by extracting and evaluating their internal features. The results provide core technical guidance and key supporting data for authenticating webcam videos within the domain of forensic science.

The rapid advancement of computer image production technology presents a grave peril to the trustworthiness of digital images, necessitating a great practical requirement for research on computer generated image detection technology in the realms of digital forensics and judicial assessment. Therefore, we propose a deep local binary pattern network (DLBPNet) to detect computer-generated image (CGI). Specifically, we first designed a deep local binary pattern module, which has three parallel branches, each of which utilizes a 1 × 1 convolution layer to learn the correlation between color channels, a learnable pre-processing filter to eliminate information redundancy, and an LBP submodule to extract low-level discriminative features. Then we feed the output of this module into successive generalized central difference convolution modules to further learn the higher-level hierarchical representation for making decision. Our proposed DLBP-Net network was confirmed to be effective in both detection accuracy and generalization ability through extensive experiments, which yielded detection accuracy of 94.35% on SPL2018 dataset, 94.03% on DSToK dataset, and 93.87% on the mixed dataset.

In the process of controllable source audio frequency magnetotelluric measurement data processing, selecting the same set of parameters under different geological and geophysical backgrounds will cause large errors in the inversion results, and different parameters under the same background will also have a large impact on the inversion results. Through the collection of geological and physical data, the author established a forward model, and carried out experiments on different inversion parameters for the forward data. Finally, the optimal inversion parameters were selected, which improved the inversion accuracy and achieved good results, and summarized the data processing process.

Lonicera japonica is a traditional Chinese herbal medicine, and root rot is a common systemic disease on Lonicera japonica, and the occurrence of root rot seriously affects the yield and quality of Lonicera japonica. In this paper, we study to realize the detection of root rot plants in a wide range of Lonicera japonica fields based on UAV remote sensing images to help accurately detect and treat root rot plants. A YOLOX target detection model was trained using UAV orthophotos to identify root rot disease plants and healthy plants in the field. After training and optimization, the average precision (Average Precision, AP) of the model reached 91.75% and 94.44%, respectively, which can efficiently and accurately detect root rot plants in the field for timely treatment.

Considering the impact of background noise, the traditional bridge crack detection method cannot completely extract the crack features and effectively fuse them, resulting in low crack detection accuracy, false and missed detection. This paper presents a bridge crack detection algorithm based on improved U-Net (I-U-Net) network, specifically designed to address the mentioned challenges. Firstly, the residuals module with bottleneck structure (Res-Bot) is combined with the U-Net model to obtain complete crack information. Secondly, in order to eliminate the interference of background noise, optimized convolutional block attention module (O-CBAM) is introduced, adding dimensionality reduction convolutional layer and dilated convolutional layer, which effectively reduces the number of network parameters while simultaneously improving the feature extraction capabilities of the network. Thirdly, based on the idea of multi-size fusion, different sizes of convolution kernels are selected according to different levels to extract crack information, and the encoder-decoder-multi-size fusion (EDMSF) module is constructed. Finally, the effectiveness of the I-U-Net network is demonstrated through the public bridge crack dataset and the self-made dataset.

Anomalous traffic detection in the network is one of the essential components of network security protection. Neural networks are widely used in intrusion detection systems, which can learn the ability to distinguish between regular and attack traffic in the network through training. However, most of the existing methods do unimodal anomaly detection on temporal features. Based on this, our paper proposes a hybrid anomaly detection model based on Auto-Encoder, which uses Damped Incremental Statistics to extract features from the data to obtain Time-domain features, then uses Discrete Wavelet Transform to obtain Frequency-domain features of corresponding dimensions using two sets of Auto-Encoders to generate the RMSE in two modules and weighted sum to get the final RMSE value. The anomaly detection problem is transformed into a refactoring problem of the time series in two spaces. The classification algorithm compares the anomaly with the threshold value, thus determining whether it is an attacking traffic. The experimental results in UNSW-NB15, IDS2017, and IDS2018 datasets show that the method in paper is better than the traditional unimodal anomaly detection method, which improves 8.11% and 7.31% in Precision and Recall, respectively, and the Precision can be up to 99.99%, which proves the validity of the method in our paper.

Due to their high reliability, high security, and low-cost characteristics, one-time programmable (OTP) memories are widely used on many occasions where stored data does not need to be reprogrammed on-site. However, the great difference between the interface protocols of mainstream OTP memories on the market and the Advanced Microcontroller Bus Architecture (AMBA) bus protocol commonly used in SoC systems makes it impossible to integrate the memories directly into the chip. To eliminate this limitation, this paper designs an OTP memory controller to realize the conversion function of OTP memory interface protocol and Advanced Peripheral Bus (APB) protocol, so that CPU, Direct Memory Access (DMA), or other modules can access the memory through the APB bus protocol. Through simulation, the results show that the controller functions well in controlling OTP memory to power-up, program, read, read in test mode, and power-down.

SRv6 has been constrained in its deployment in a large-scale operator network since the lengthy packet header decreases the forwarding performance in network processors. In this paper, we analyze in depth and reveal the limiting factors on network processor performance introduced by the long SRv6 packet header. The required resources for processing SRv6 and G-SRv6 packet headers are also compared in terms of cache space, processing bandwidth and routing table in network processors. By constructing an experimental network to measure the throughput and forwarding latency of network processors when handling SRv6 and G-SRv6 packets, the experimental results show that G-SRv6 can improve the packet throughput of network processors and reduce packet forwarding latency, which is beneficial for the largescale deployment of SRv6. This paper provides valuable insights and guidance for the design of network processor architectures targeted at G-SRv6 applications.

A parameter sharing model using syntax based graph convolutional neural networks to capture text structure information is proposed to address issues such as error propagation and ignoring inherent relationships between subtasks in pipeline models. This article will specifically introduce a model that combines parameter sharing mode, including the motivation for designing the model, special annotation strategies, and model structure, experimental settings, and analysis of experimental results. Traditional entity relationship extraction was composed of two sub tasks: entity recognition and relationship extraction. When there are errors in entity recognition, the error information will be propagated to the relationship extraction task, leading to error accumulation. In response to this issue, this article proposed a study on joint entity relationship extraction based on graph convolutional neural networks.

Physically Unclonable Functions have gained significant attention in recent years for their ability to address hardware security concerns. This work presents a method to implement ring oscillator-based PUF on the FPGA platform. Specifically, an 11-order oscillation ring is configured by cascading inverters, exploiting the little variations in the FPGA manufacturing process to generate unpredictable, random output. To evaluate the proposed structure, experiments were conducted using the Xilinx Kintex-7 FPGA chip. The experimental results demonstrate that the designed RO-PUF exhibits a uniformity of 47.30%, uniqueness of 47.46%, and reliability of 97.07%. It possesses several advantages, including ease of implementation, minimal resource utilization, and the elimination of manual layout and routing requirements.

Individual causal effect (ICE) plays an increasingly important role in personalized medicine and government decision-making. However, previous ICE focuses on the conditional treatment effect and the individual characteristic is embodied by the features, which makes it impossible to obtain the individual information the features are unable to include. To address this issue, we propose a heterogeneous regression model that includes a fixed individual effect and develop a ridge-fused penalty-based criterion to estimate the individual parameters. By deriving the asymptotic distribution of the estimator of individual causal effect, we make an effect individual inference for ICE. Simulation studies pose supportive evidence that the proposed unified structure of individual analysis including estimation, and inference performs well with finite samples, and a real data example is provided to illustrate the flexibility of ICE.

In this paper, STM 32 single-chip microcomputer is the control core with environmental temperature and humidity sensor environment sensor, soil humidity sensor and light sensitivity sensor to realize the real-time detection of environmental temperature and humidity, soil humidity and light intensity of potted plants. According to the current light intensity, the intelligent rotating shading plate shades or fills the light of potted plants, and the water pump is automatically opened according to the soil water content. At the same time, an OLED display is included, which is convenient for users to more conveniently and intuitively understand the plant growth in the pot. In addition, the intelligent pot system has a Bluetooth communication function, which can remotely send instructions on the mobile phone to operate the pot to perform shading or watering action. This intelligent flowerpot system provides a safe and comfortable growth environment for potted plants and brings a more intelligent and convenient use experience for growers.

In Wireless Sensor Networks (WSNs) composed of sensor nodes with location mobility, most wireless sensor nodes use batteries as their power and have limited energy, so designing energy-conserving routing protocols is crucial for the long-term reliable operation of mobile WSNs. SPIN-PP is a simple energy-conserving plat routing protocol, however, it is for fixed sensor nodes. To satisfy the requirement of saving energy for mobile WSN, this study is carried out. Aiming at the situation of random movement of sensor nodes, based on the idea of plat routing protocol, this paper adopts the strategy of mobile sink node (sink) and uses the method of vector projection to obtain the optimal trajectory of the sink node, so as to decrease the total energy consumption of communication. The simulation results show that, compared with SPIN-PP protocol, the proposed method has great advantages in conserving energy and node runtime in a period.

This paper deeply discusses the basic knowledge of analog integrated circuit design, focusing on the physical principle and working principle of MOS (metal-oxide semiconductor) devices, as well as the design principles of single-stage amplifier, differential amplifier and current mirror. By understanding these basic concepts, we can better understand the core idea of analog integrated circuits. In addition, this paper also describes the circuit performance analysis tools in detail, including the basic use of Multisim, the establishment of quantitative analysis tools and the interpretation and application of simulation results. These tools can help engineers quickly evaluate the performance of the circuit, optimize the circuit design, and improve the reliability and stability of the circuit. Finally, this paper shows the application of analog integrated circuits in interactive intelligent wireless charging system. Through the use of analog integrated circuits, the intelligent control of wireless charging equipment can be realized and the charging efficiency can be optimized, which provides strong support for solving practical problems.

Future war scenarios will develop in the direction of systematization, networking, and intelligence, present data links are equipped with different frame formats, communication protocols, and application scenarios, resulting in the formation of “information islands” for each data link, which cannot achieve interconnection and information sharing between different platforms. In response to the above problems and the urgent need for cross-platform interconnection and interoperability, this paper proposes an IP-based integrated aircraft network architecture for heterogeneous data links, which builds a unified interconnection network based on IP hierarchical structure through the slicing encapsulation and reassembly technology of IP packets. The architecture shields the differences of the underlying data links and achieves the transparent transmission of IP packets, thus improving the connectivity of present data links. It is an important support for the in-depth information sharing of different communication platforms in the future.

Traditional bridge cable stay detection has problems such as difficulty in detection and weak detection capabilities. This article studies the detection performance of intelligent detection robots on cable-stayed cables through experimental tests. The research results summarize the analysis and evaluation of the experimental data and point out that the intelligent detection robot combined with the proposed detection method has achieved significant improvements in indicators such as accuracy, recall rate, precision rate and F1 value. Experimental results show that the method is able to accurately detect defects in stay cables and provide detailed information about the type and location of defects. The results of this research provide an innovative solution for bridge maintenance and safety management and have broad application prospects. Future research can further optimize the design and algorithm of the intelligent inspection robot, expand its application in the field of bridge structure inspection, and combine big data and artificial intelligence technologies to further improve the accuracy and intelligence of bridge cable stay inspection.