This PDF file contains the front matter associated with SPIE Proceedings Volume 12506 including the Title Page, Copyright information, Table of Contents, and Conference Committee Page.

The study of ancient Chinese writing has great cultural and historical value. Text annotation is a time-consuming and laborintensive part of ancient writing research. In this paper, we construct a deep learning model for ancient Chinese text detection which combines Swin Transformer and Mask R-CNN. To solve the overlapping detection problem, we propose Text Non-Maximum Suppression (text-NMS) and text-NMS loss. The former is to weed out redundant subtext bounding boxes in the Non-Maximum Suppression process, and the latter further rectifies the detection failure missed by text-NMS in bounding box regression. Experiments on the Chinese Stone Inscription Dataset show that the proposed algorithm can improve the accuracy of ancient text detection. The text-NMS and text-NMS loss algorithm boost the average precision (AP) of Swin-small and Mask R-CNN from 64.4% to 65.8% with few additional hyper-parameter and computational overhead.

In order to prolong network life, improve network throughput and reduce energy consumption within the network of WSN, a multi-hop routing algorithm for WSN based on non-uniform clustering is proposed in this paper. This protocol effectively alleviates the problem that in the multi-hop data transmission network, the cluster head close to the base station bears too many tasks, and the energy consumption is high, which leads to the premature death of the node. The cluster heads within the direct communication radius in the network will communicate directly with the base station, and the other cluster heads will select the neighbor cluster heads with the least communication cost as relay nodes for multi-hop transmission. Finally, the simulation results show that the protocol can effectively alleviate the premature aging phenomenon in the multi-hop network, and the energy consumption of the nodes in the system is balanced, which meets the design requirements of the wireless sensor network routing protocol.

Images can convey information and needs to be protected, which is usually done by encryption. Traditional image encryption methods usually use software or hardware encryption, which has improvement approach in terms of better safety and robustness. And the study arouses sympathy to researchers based on both software and hardware encryption. This paper proposes an encryption method that combines software algorithm encryption and hardware encryption using a spatial light modulation (SLM) and 4f optical system. Firstly, the image is chaotically encrypted using the chaos encryption algorithm. Secondly, the discrete cosine transform (DCT) is adopted for further encryption based on chaos encryption picture. Thirdly, the cipher text image is transformed into the 4f system using SLM, which performs a random amplitude on the image to achieve hardware encryption. Finally, the simulation and experiment are performed to verify that the scheme, which shows that the method can realize the encryption and unencryption under the combination of software and hardware encryption. The proposed method has application value to guarantee the higher security of information.

Speaker embedding is a state-of-the-art front-end module, which is used to extract discriminative speaker features for speaker-related tasks. The Time Delay Neural Network (TDNN) has been a classical network architecture since it was first applied on speaker related tasks known as X-vector. In this paper, we propose new network structures based on current popular ECAPA-TDNN. We propose a dynamic kernel convolution module to extract features from short-term and long-term context adaptively, thus achieving multi-scale receptive fields. We also apply three enhanced attention modules instead of plain Squeeze-Excitation (SE) layer to realize more efficient information interaction between channels and spaces. The proposed architectures are superior to the most advanced network, with an optimal Equal Error Rate (EER) of 6.40% and a parameters reduction of 6.32%, they also achieve better performances when speaker utterances are shortened.

At present, the layout of sprinkler position in the irrigation system mainly depends on manual design. When the irrigation scale is large, the manual layout is difficult to realize, and the unreasonable layout is easy to cause problems such as sprinkler irrigation equipment and water resources waste. In this paper, the k-means algorithm in machine learning is applied to agricultural irrigation, and the concept of intelligent sprinkler irrigation layout design is proposed. Firstly, the sprinkler irrigation area is simulated and stored in the computer based on the inner and outer point discrimination algorithm of polygon, and then the irrigation area is clustered and divided based on the k-means algorithm to obtain each cluster. The central point of each cluster is the best location for sprinkler installation. In the comparative experiment of irrigation areas with different shapes, it is concluded that the irrigation coverage and irrigation uniformity coefficient have a good performance. Compared with manual design, the intelligent irrigation layout proposed in this paper is more intelligent and reasonable.

This article studies the multi-target classification issues of multiple-input multiple-output (MIMO) radar, a complexvalued back-propagation (CV-BP) network structure with the cross-entropy cost function is designed. The complexvalued echo signals captured by MIMO radar are pre-processed to provide more data for the classification, and then the pre-processed complex-valued data is directly selected as the recognition feature, combined with the CV-BP network to classify multiple targets. This scheme can not only effectively avoid the process of target feature extraction manually, but also improve the accuracy of classification by utilizing the amplitude and phase information of the target. The simulation results demonstrate the effectiveness and robustness of the CV-BP scheme under different SNRs.

To solve the problems that ant colony algorithm often gets sub optimal solution, the turning angle is too large and the convergence effect is poor when planning the global path of the robot, the traditional ant colony algorithm is improved. The grid map is applied to construct the working surrounding, the path heuristic function is designed to enhance the searching ability and convergence effect of the algorithm. Considering the influence of turning Angle on travel time, the heuristic function of turning Angle is introduced to ensure the ant colony to have more guidance in the planning and make the path smoother. Pheromone updating rules and path evaluation mechanism are established to guide ants to approach the path with the best comprehensive index. The algorithm is simulated on the MATLAB platform, the simulation indicates that the optimized algorithm has good convergence effect and strong optimization ability.

In today’s era, information technology is constantly updated, and the degree of social information is getting higher and higher. Power grid enterprises have a large amount of asset data, and There are many types of assets, such as tools, information system and vehicles. These assets need to be managed. Although there are unified codes for the types of assets at present, manual input is still inaccurate, so there will be some inaccurate descriptions, which makes it very difficult to promote the management of power grid assets. In this paper, an improved FastText text classification method is proposed to identify the information of various power grid assets. The purpose is to ensure that the technical object type coding of power grid assets matches the type of power grid assets, and to realize the efficient quality management of power grid asset object type coding data Concrete by using n-gram model information, read between the word and the word order in hidden layer of network by traditional method and the average improvement for the average sum of squares, and the network output layer for the softmax improved hierarchical softmax and negative sampling method, improved rapid extraction and classification of information, and can better meet the automatic identification data grid assets. The experimental study in this paper proves that the classification accuracy of the improved FastText model on the grid asset data set is significantly improved, and the intelligent classification of grid assets can be realized.

The Internet and digital technology have impacted all traditional content industries, and the music industry is naturally inevitable. Therefore, research on the Internet, music industry, and digital music copyright has become the focus of academic circles in various countries today. The purpose of this paper is to study the application of copyright protection of musical works based on digital audio watermarking algorithm. Firstly, the theoretical basis of digital music copyright and discrete wavelet transform is expounded. The security of the audio watermarking system is improved by using the pure sequence modulation; the watermark is embedded in the double transform domain according to the energy gathering effect of the discrete cosine transform, and finally the blind extraction of the watermark is realized. Finally, the two algorithms are compared and discussed. The experimental results show that, under the same attack conditions, the normalized correlation coefficient between the original watermark and the extracted watermark of the watermarking algorithm based on pure modulation and double transform domain is larger. The advantages of the watermarking algorithm based on mixed pure modulation and double transform domain are analyzed, which makes the application of wavelet transform coding in music copyright information more practical and extensive.

This paper presents an Unmanned Surface Vehicle (USV) path planning algorithm based on a Clockwork-RNN (CW-RNN) framework. Compared with RNN, CW-RNN divides the hidden state matrix into several small modules and uses a method similar to clock frequency mask to divide the memory of RNN into several parts, so that each part of CW-RNN memory matrix can process data at different times and enhance the memory effect. The existence of ocean current will cause drifting motion in the navigation track of unmanned ship, and make the ship deviate from the planned path and direction. In view of this interference factor, ocean current vectors of different sizes and directions are added in the simulation environment to make the environmental model closer to the actual sea surface. As the environment becomes more complex, reinforcement learning takes a long time to train in the complex environment and is not easy to converge. Therefore, this paper combines reinforcement learning method with the traditional path planning method Dijkstra algorithm, inputs the local map information detected by unmanned ship into Dijkstra algorithm to give task sub targets, and uses these sub targets to guide unmanned ship and improve the search efficiency of neural network. Finally, the simulation results and analysis show that the training algorithm is effective in the current environment.

The stability control of quadrotor UAV system is a challenging problem due to its nonlinearity, strong coupling and underactuation characteristics. In this paper, a fractional order PID control strategy (FOPID) is proposed to realize the high-precision control of UAV system by utilizing the multi-degree of freedom characteristic of FOPID, SA-PSO is used to tune the parameters of FOPID. Compared with the standard PSO optimization process, the convergence speed is faster and the accuracy is higher. Simulation results show that the proposed PSO parameter optimization based on SA algorithm makes FOPID controller have better performance than traditional integer order PID controller.

As a brain-inspired artificial neural network computational model, a recurrent spiking neural network is composed of biologically plausible spiking neurons, which has taken on increasing importance in this study mainly include complex network structure and implicit nonlinear mechanism. This paper presents a learning algorithm with synaptic delay-weight plasticity for recurrent spiking neural network, which may enable real-time communication of complex spatiotemporal spike trains simulating organisms. First, by copying the hidden layer between the input and output layers, a context layer can be created. A total error for spatiotemporal pattern is then introduced, along with the learning rule for synaptic weights with synaptic delay plasticity. Moreover, by using synaptic delays of spike train from postsynaptic neuron, the proposed algorithm defines the learning rules for synaptic weights. In addition, the performance of the learning algorithm has been successfully tested to achieve high-precision learning with limited iterative resources. Finally, the main factors have been evaluated qualitatively and quantitatively, such as different lengths and frequencies for desired output spike trains.

In this paper, we constructed a one-dimensional convolutional neural network as a classifier model for spatial object classification. Considering that there are few available training samples obtained from actual measurement, combining with the characteristics of actual measurement data, we simulated a large amount of data for training and testing. The simulation results show that our method has a high classification accuracy and can overcome the problems existing in actual measurement, such as tracking mixed batches to a certain extent, and it can also effectively solve the problem that it is difficult to directly train neural networks because of the small number of spatial target samples, which take advantage of neural network autonomous learning and memory to reliably identify features.

Forest canopy structure is very important to measure forest change and forest coverage. And TerraSAR-X data are well suited for inversion applications at tree height. Based on the Random Volume over Ground model, the three-stage algorithm and its PSO improvement are studied in this paper. Taking the TerraSAR-X data of Mengla County in Yunnan Province China as the data source, the forest height inversion algorithm were compared in the experiment part. Finally, the results are verified with the field measured data. The results show that the precision of forest height inversion based on the PSO intelligent algorithm is better than the traditional three-stage algorithm, and the correlation coefficient is improved by more than 20%.

LiDAR-based SLAM Systems are widely applied in robotics for their accuracy and robustness. However, accurate localization in small-scale indoor scenes is challenging since the point cloud features of the sparse line scan LiDAR cannot always provide sufficient space constraints. In our experiments, even state-of-the-art methods have heavy odometry drift. In this paper, to address this problem, we propose a method that can improve the performance of existing algorithms in small-scale indoor scenes. By installing a LiDAR perpendicular to the existing LiDAR, our method enhances the constraint in the vertical direction. We test our method on LOAM and FAST-LIO2, and the results show significant improvements on our own collected datasets. In addition, to accurately merge the two LiDAR’s cloud points, we propose a high-accuracy dual-LiDAR calibration method with rotation and translation errors less than 0.005 rad and 0.01 m respectively.

An improved grey wolf optimization (IGWO) is proposed for the travel salesman problem. Based on traditional grey wolf optimization, a genetic gene sequence was introduced for the first time to initialize the population to improve population quality. Subsequently, the cosine adaptive function is used to balance the global and local search abilities. Finally, a distance heuristic factor is introduced into the update strategy to enhance the local search capability and improve the robustness of the algorithm. Several cases were randomly selected from the TSPLIB database for the simulation experiments, and it was found that the IGWO has higher stability and accuracy under the same conditions.

Gliomas comprise around 80 percent of all malignant brain tumors which can be further classified into the low-grade glioma and high-grade glioma categories. Compared to low-grade gliomas, high-grade gliomas show more malignant behavior since they usually grow rapidly and frequently destroy healthy brain tissue. Therefore, it is important to determine the malignancy of gliomas for initial treatment plan. Most existing methods are based on radiomics or transfer learning which extract features from single slice without considering that 3D features between adjacent slices can provide stronger discriminative power. In this work, we propose to incorporate the attention mechanism and pseudo-3d module into a deep convolutional network architecture, which can reduce model size and produce fine features for glioma grading. Our work focuses on the inter-slice relationship and propose an attention unit, named “Inter-Slice Attention Module”, which adaptively refines intermediate feature maps by modelling dependencies between adjacent slices. We evaluate our method on an open dataset of gliomas, achieving mean accuracy of 89.47%.

A naval vessel made of ferromagnetic materials will generate magnetic anomalies under the influence of the geomagnetic field. This situation will affect the safety of the naval vessels, so it is necessary to be minimized. The position and number of degaussing coils in the traditional degaussing system are relatively fixed after the ship is built, so it is difficult to adapt to different magnetic anomalies in different areas. In addition, the traditional degaussing system optimizes the coil current only to minimize the residual magnetic field peak, without considering the magnetic field gradient. Aiming at this situation, this paper proposes a multi-objective optimal design scheme of ship degaussing coil based on high-dimensional variable selection technology and multi-objective solution technology. In this paper, LARS technology is used to reduce the complexity of the problem, and MPSO technology is used to solve the multi-objective optimization problems. Experiments show that this method cannot only rank the importance of degaussing coils, but also effectively reduce naval vessels magnetic abnormal signals, which has practical guiding significance.

In order to solve the problem of the order picking optimization in warehouse line, this study puts forward the concept of class TSP (Travelling Salesman Problem) based on a model using graph theory knowledge modelling. With the aid of the thought of classical TSP, it puts forward three principles, based on a fixed starting point (not repeatable) integer programming model; The solution of the model is also based on a simple optimization algorithm (vertex translocation method) of the classical TSP, and the appropriate modification is made. For the problem of unfixed starting point, the global traversal algorithm is added. Through the reconstruction of the line sequence, a TSP-like problem is successfully transformed into a TSP, and a model is finally established, which is a prominent innovation point of this paper. The calculation method of adjacency matrix W is given in the postscript.

Biological sequence alignment is one of the most basic research topics in bioinformatics. The double sequence alignment algorithm based on dynamic programming mainly uses iterative algorithm and vacancy penalty rules to compare gene sequences one by one, calculate their similarity scores, and finally obtain the best alignment between sequences through backtracking analysis. This paper studies the analysis and implementation of the global double sequence alignment optimization algorithm based on dynamic programming.

In private cloud environments, resource usage is determined based on experience, which often leads to excessive resource allocation and low resource utilization. This paper proposes a private cloud resource consumption prediction algorithm based on combinatorial optimization algorithm. Based on the virtual machine performance data collected from the private cloud platform, the key features are selected by recursive feature elimination method (SVM-RFE), and then the model is trained by eXtrem Gradient Boosting (XGBoost) algorithm for model training to predict the resource usage. Compared with Random Forest, LSTM and other algorithms, the proposed algorithm has higher prediction accuracy and smaller prediction error. This paper adopts a data-driven approach to achieve intelligent prediction of resource usage in private cloud environments, which improves resource utilization and provides decision support for optimal allocation of resources.

For monitoring of cutting-edge technologies by obtaining the massive data on the internet, the pyspider framework is used to regularly crawl the information of a large number of websites, and the crawled data is regularly imported into the business application system through python scripts. At the initial stage of the project, the distributed architecture is used as officially recommended by pyspider. Later, it is optimized as the clustered architecture in order to adapt to the actual network environment and the characteristics of crawler tasks. After testing, the work efficiency and stability of the improved architecture have been greatly improved and the expected results have been achieved.

Short Text Classification is the fundamental task in the nature language processing. There is a lack of language structure and uneven classification of data samples in short texts, which limit the development of deep learning based short text classification. To address the limitations of text sequences, we propose using a large-scale pre-trained language model Bert to obtain feature information between words and bureaus in the text, Graph Convolutional Network (GCN) with double-layer convolutional network can obtain the dependency relationships between words. We propose to combine Bert with GCN in short Chinese medical texts, where BertGCN outperforms better than other’s methods in classification accuracy.

According to the coverage of cell phone signals in geographic space, combined with time series of cell phone, positioning data can restore the complete realistic activity trajectory of the population, so as to obtain the characteristic information of spatial distribution and activity connection of the population. In this paper, the user location dataset obtained from CMCC is pre-processed, then mined by hierarchical clustering algorithm and combined with data visualization techniques. The experimental results can provide decision support for commercial promotion, so as to carry out targeted business layout.

This paper presents a ballistocardiogram (BCG) classification method based on 1-D convolution neural networks (CNNs), which provides an auxiliary basis for clinical diagnosis, especially in the monitoring of cardiac function in the elderly. Four categories of BCG signals are used as input of 1-D CNN. By moving the convolution kernel on the time axis, the temporal variability of BCG signals can be better satisfied while retaining the frequency band correlation. Then they were sent to the multi-layer sensor, after that the features processed by softmax classifier were classified. We obtained an accuracy of 93.39%, of which the H class is 95.36%, the C class is 86.19%, the D class is 95.31%, and the Y class is 96.57%. Compared with the existing research results, the proposed method achieves a superior classification performance. This method is simple, fast and highly generic, which can serve as a reliable adjunct tool for the daily diagnosis of heart disease in the elderly.

In some long and untrimmed videos, locating the important and key segments can be a very challenging task for temporal action detection. Current methods make remarkable progress when it comes to RGB images. The aim of this paper is to develop a method with the assistance of dynamic model of human body skeletons to address this problem. To this end, we propose a Feature Pyramid Graph Convolutional Network (FP-GCN). The introduced model contains a Feature Encoding Module to encode skeleton data with graph convolutional networks, a Feature Pyramid Module to exploit the inherent pyramidal hierarchy and an Action Detection Module to generate the final prediction results of the detection. We experiment our approach on NTU RGB+D and THUMOS14 datasets and obtain a satisfactory result.

Due to the complexity of the battlefield environment at sea and the diversity of attack forms, a single fixed maintenance strategy cannot cope with the varied attack modes of the enemy at all times after the ship system is attacked. Aiming at the marine battle environment, this paper simulates the multi-wave attack chain and ship defence system, so that the ship equipment node has different wartime importance according to different combat environment, combined with the explosion and damage model, interception and confrontation model. The repair sequence of damaged nodes under multiple constraints was abstracted to the path planning problem similar to MSTP, and an Attack Defence Recovery Strategy (ADRS) was proposed by combining adaptive genetic algorithm with the wartime importance of equipment nodes. Experiments simulation has been taken under the same constraints and multi-wave enemy attack with different recovery strategy system of confrontation and resilience recovery. The results show the system resilience recovery based on ADRS is effective under the waves of attacks, reduces system damage, improves the system robustness, and especially greatly increases under relatively severer damage.

In this paper, an improved ant colony algorithm is proposed to solve the transportation optimal path problem. The algorithm can quickly find the optimal path by improving heuristic function, pheromone and neighbourhood-search. Experimental simulation shows that the improved ant colony algorithm proposed in this paper has certain advantages over genetic algorithm and ant colony algorithm. It not only obtains the minimum value of the optimal path length, but also has the fastest convergence speed, which effectively solves the problem of optimal transportation path planning. The improved algorithm effectively accelerates the transportation speed, shortens the transportation time and saves the transportation cost.

With the rapid development of Internet, data has become increasingly important in today’s society. With the widespread use of data, the issues in data use like low efficiency, high cost, lack of supervision might hinder the development of enterprises and companies. Therefore, the way to mine data resources as a new development power is extremely important. In this paper we applied the convolutional neural network in image for testing. We used LDA, XGBoost and other models, based on a large number of questionnaires questionnaire data. In the process of data processing, we fitted the 3D data after Emmett interpolation to make the data more intuitive. The analysis of the model shows that the fit of the regression curve, R², is 0.173 and Vif is less than 10, indicating that data has become an important driver of socioeconomic development, both from a macro and micro perspective.

Multiple object tracking (MOT) has become a very important task in the field of rail transit. Recently, the most effective method in MOT tasks is tracking by detection. Because there are many pedestrians under a single camera in the railway station, the occluded pedestrians will generate detection boxes with low-score. Once these low-score bounding boxes are mistakenly classified as background by the detector, the traditional tracking model may discard these boxes, resulting in poor tracking effect. In order to solve this problem effectively, this paper proposes a novel model with Generic Association Mechanism (GAM), which can pay attention to every detection box instead of only focusing on the highscore ones. This method utilizes the similarity between these low-score detection boxes and trajectories to restore true objects and filter out the background detection boxes. More attention can be focused on these low-score detection boxes through this method, which contribute the most to the tracking effect. This method can efficiently avoid the loss of true objects and fragmented trajectories of occluded objects. Finally, we test this proposed model with other models on several public datasets, and the results show that the proposed method achieves more significant performance improvement.

Aiming at the problem that the position of skeleton obtained by the method based on topology refinement is not accurate enough, and the connectivity of the skeleton obtained by distance transform cannot be guaranteed, this paper combines the advantages of the topology refinement algorithm and distance transformation algorithm to propose an improved skeleton extraction algorithm. This method first obtains the skeleton points through distance transform, and then obtains the initial skeleton through the refinement method; last uses the skeleton points obtained by distance transformation to screen each branch obtained through the refinement method to ensure the accuracy of the skeleton in the later applications. Experiments on the MPEG7 dataset demonstrate the effectiveness of the algorithm.

With the improvement of facility condition and network speed, the network traffic has also presented an exponential growth in recent years. However, a growing number of problems concerning cyber security have appeared. Anomaly network traffic detection can identify abnormal traffic from massive traffic. Accurate identification can reduce anomaly network traffic and protect user client. Our research is based on the traffic collected and processed by National Chiao Tung University. These data include normal traffic and abnormal traffic, the former one being majority. We propose a processing method with high accuracy. We first pre-sampled the data set, and then analyzed the data features. Later on, based on theoretical research, we integrated the model of random forests and other boosting method and proposed a greedy multi-classification to binary classification model based on ensemble learning model.

Intelligence inspection of the electrical power grid can effectively improve the efficiency of inspection. However, the complex scenes between grid work and the interference of similar objects seriously affect the recognition accuracy of target identification. To address the above problems, this paper proposes the detection of abnormal dangerous behaviors in power grids based on contextual feature fusion. First, the feature extraction backbone network is constructed to obtain image features. Second, the hierarchical contextual attention mechanism is constructed to capture contextual features. Finally, the target detection model with contextual feature fusion is constructed to achieve grid abnormal risk behavior detection. The model proposed in this article is compared with the existing object recognition model in the simulation of the three datasets of Safety helmet (hardhat) wearing detect dataset, Hard Hat Workers dataset, Safety Helmet Detection dataset. A large number of experiments have proved the object recognition algorithm proposed in this paper is effective and outperforms existing algorithms. The average recognition accuracy of the proposed model is 0.948, which is improved by 1.02%.

The teaching theory is disjointed from social needs in the computer professional courses. To solve this problem, we propose a new model that incorporates the Aliyun platform and large-scale open online course system. MOOC are a useful addition to theoretical teaching. The Aliyun platform has experimental teaching with real cases, which alleviates the environment configuration of students and increases their interest in learning. After a year of operation, there has been a significant increase in student experimentation. Statistical results show that the students’ final exam scores increased by 12.50% and their operational scores by 14.80%. The proposed model is more suitable for modern computer teaching in cloud environments.

Multiple Independent Levels of Security (MILS) is widely used in the design of high assurance operating system. By separating the system into components, and making the components run in different domains, the kernel can control and monitor information flow between components to enhance the security and availability of system. However, due to the complexity and certification cost issue associate with large monolithic kernel, MILS architecture is mainly used in microkernel system. But we still want to use the idea of MILS in monolithic kernel system to improve the security. In the Linux, although there are some access control models based on the concept of domain (like SELinux). Limited by the feature of shared kernel, the security of system is affected by the vulnerabilities in itself. Therefore, this paper proposes a scheme of constructing multiple independent isolated domains based on virtualization technology in Linux. We developed on Linux kernel and QEMU/KVM hypervisor, exploiting the isolation feature brought by virtualization to achieve data isolation. We build domain from virtual machine, so that we can separate origin system into components and run them in domains. In the host, we take control of all domains and implements a secure communication mechanism between domains. By using this secure channel, we can monitor the data transmission between domains, and control the information flow according to the security level of the domain. Finally, we evaluated the effectiveness and efficiency of our communication mechanism.

In this thesis, we study the cognitive cooperative non-orthogonal multiple access network with interference power constraint in underlay mode, where a primary receiver is situated at the communication range of the secondary network. In the cognitive secondary network, the secondary originating node transmits information with the cognitive close user directly and with the cognitive remote user by the aid of multiple relays under cognitive radio constraint. Secondary originating node sends the mixed message to the cognitive close user and to multiple relays via NOMA principle. Through the signal to interference plus noise ratio between secondary originating node and multiple relays, the best relay is opted to forward the decode signal to the cognitive remote user. In order to gauge the performance of the system accurately, the exact closed form formulas for the outage probabilities of the cognitive close user and the cognitive remote user are deduced respectively over Rayleigh fading channels. Experiment results indicate that power allocation has a great influence on the performance of system for NOMA network and it is an effective way for improving the performance of cognitive cooperative NOMA network that increasing the quantity of relays.

In order to enable Internet of vehicle users to discover the surrounding available wireless networks in time and reduce the energy consumption caused by network scanning in a heterogeneous network environment, this paper proposes a vertical handover algorithm based on dynamic scanning period and grey prediction method. In the network discovery phase of vertical handover, this method dynamically changes the scanning period according to the speed of the vehicle, so as to achieve a balance between the activate times of the interface and the timely discovery of the network. In the network selection phase, this paper uses the grey prediction algorithm to predict received signal strength of the current access network, so that Internet of vehicles users can make decisions and switch in advance. The simulation results show that compared with existing algorithms, dynamic scanning period method proposed in this paper can discover the network in time and reduce the energy consumption of the terminal. The use of grey prediction algorithm enables Internet of vehicles users to effectively predict the received signal strength and make handover decisions more timely, and the handover call drop rate is improved.

Aiming at the fact that the current graph convolution operation based on skeleton graph is limited in local adjacent nodes, or the overall relative position information of skeleton is omitted, an enhancement method of joint point position information based on relative position encoding of skeleton is proposed. The proposed method takes the central joint point of the human trunk as the root node, and all joint nodes form a tree structure according to the natural connection of the body, and the code of each joint node inherits the code of its parent node and also includes its own number in the sibling node. In addition, considering that the number of channels in the graph-based convolutional network model is generally larger, and the channel information itself has a strong correlation, the channel information frequency division and recombination operation is proposed to reflect the difference of information in different frequency bands in the channel. Experiments show that the proposed method has a certain effect on improving the effect of the embedded model.

Network flow watermarking has been employed to analyze whether two flows are correlated, which is often applied in non-cooperative networks. The detection performance of network flow watermarking is poor due to delay jitter, packets loss etc. In this paper, we improve the detection performance of network flow watermarking by using Luby Transform (LT) codes, which can take full advantage of network flow by automatically adapting to the length of network flow. To further improve the performance of LT codes for network flow watermarking, an optimal degree distribution is generated through a heuristic algorithm. Theoretical analysis and simulations show that our proposed method outperforms others in terms of the detection performance.

As a learning goal of deep neural networks, time-frequency masking has become research focus of supervised speech reconstruction. The previous study proves that the complex-value ratio mask (cRM) can simultaneously estimate the amplitude and phase components of the clean signal in the noisy speech. Compared with the other time-frequency masking features, the best performance can be achieved. However, because the imaginary structure is not obvious and the neural network learning is difficult, there is still no accurate estimation method at present. In this paper, we improved the speech reconstruction with the complex-valued full convolution neural network (CFCCN). Based on the theoretical of complex-value neural network, the complex-valued building blocks are designed for CFCCN to handle complex domain operations and estimate cRM. The building blocks include the complex convolution filter, complex activation functions, complex batch-norm, complex pooling and the weight initialization strategies. The experiment results show that in terms of subjective and objective measurements, this work achieves at least improvement of 1.3%-12.5% in contrast to the state-of-the-arts DNN based speech reconstruction methods in challenging conditions, where the environment noises are diverse, and the signals are non-stationary.

Premature convergence and falling into local optimum easily are important problems for genetic algorithm. Therefore a new algorithm is constructed in this paper, which can be performed among multiple populations by introducing different optimization parameters. The final formation is aggregated into an elite population, where genetic operations such as selection, crossover, mutation, etc. do not exist, for the purpose of ensuring that the optimal individuals selected between different populations are not destroyed and lost. The simulation results show that the algorithm used in this paper has better optimization ability and higher calculation accuracy for solving local optimal problems of nonlinear functions.

Excel is widely used due to its continuous improvement and development on Excel function. But with the development of science and technology and the subsequent increasing data amount of to be processed, the problem that Excel still lacks of batch data processing and data verification is challenging. Based on Spring MVC + EasyUI framework, we proposed a Java J2EE IDE integrated development environment. An independent Java Development Kit was designed, which is able to be embedded into Excel. Using Json to transform the results of Java background processing into data that can be processed by the front-end, we realized the batch import and export of Excel data and data verification functions, which ensured the completeness and accuracy of the data entered while retaining the convenience of Excel.

In this paper, ant colony algorithm and genetic algorithm are combined to improve the stability and efficiency of cloud computing resource scheduling. The strategy of this algorithm is to use the local optimal solution of genetic algorithm as the initial pheromone of ant colony algorithm, and introduce load balancing adjustment factor and transfer probability into ant colony algorithm. The experimental results show that the genetic ant colony fusion algorithm in this paper has better advantages in the number of iterations, time cost, power cost and load balancing, and achieves the goal of uniform distribution of cloud computing resources.

Today, the increasingly complex external environment and the goal of building a harmonious society have put forward higher requirements for security check work. Traditional security inspection methods are facing enormous challenges. With the maturity of intelligent security technology, Intelligent security inspection has become a trend. Security check that requires both reasoning speed and accuracy, this paper proposes a lightweight target detection network, which takes YOLOX as the overall framework and Mobilenet-V3 as the backbone network. The effectiveness of the network is proved by the simulation experiment on the public data set OPIXray and can achieve a real-time computing speed of 87.7FPS.

China is a high-incidence area of earthquakes, and earthquakes cause huge economic losses and casualties every year. Earthquake disasters have adverse effects on the economic and social development of any country in our country. In the post-earthquake rescue process, the dispatch of emergency rescue teams and materials is very important, which is the basis for rescue and disaster relief, response to emergencies, and disaster relief. This paper is based on the related application research of improved particle swarm optimization algorithm. The emergency material reserve is the key to improving the emergency material support capability and an important guarantee for responding to sudden disasters. Therefore, it is necessary to select the best site selection plan for the emergency material reserve. Based on the background of earthquake disaster, this paper analyzes and studies the location problem of emergency material reserve. At the same time, in order to improve the post-earthquake emergency rescue capability, the reasonable dispatch of post-disaster emergency supplies is particularly important. The purpose of this paper is to design and implement the technology involved in the research and dispatch of earthquake emergency materials. The final result of the research shows that the required quantities of emergency materials dispatch are 9.94 tons, 6.52 tons, 4.98 tons, 7.64 tons and 5.33 tons, respectively. The corresponding emergency material dispatch shortage ratios are 6.74%, 4.33%, 3.41%, 5.17% and 4.05%, respectively. The shortage ratio of emergency material dispatch changes with the change of the required quantity of emergency material dispatch. At the same time, this paper analyzes and researches the location of emergency material reserves, in order to improve the emergency rescue capability after the earthquake and reasonably dispatch the emergency materials after the disaster, which is particularly important.

The paper aims at the problem of coverage hole affecting network service quality in WSN, a Coverage Hole Repair Algorithm based on Vertical Bisection (CHRAVB) is proposed. Three principles for selecting optimal driving node are constructed and the location of optimal target repair is determined based on vertical bisection method. Simulation experiment results indicate that the CHRAVB algorithm achieves the low redundancy and the complete repair of coverage holes with a few mobile nodes.

Aiming at the problem that frame difference method will deal with the moving target incorrectly when the target moves slowly or stops for some times, a novel method based on frame difference for moving object detection is proposed. Firstly, the algorithm confirms the number of frames used to generate background by comparing the binary frame difference of the current frame with the initial binary frame difference and establishes the background model using the determined frame; Then the background subtraction method is utilized to detect the moving object through the established background. Relevant experiments show that the algorithm can better deal with the detection of slow-moving targets.

Free space laser communication has the characteristics of large bandwidth, high speed, high security and small size, and has become the preferred way of high-speed data transmission in space today. The realization of multi-beam control in free-space optical communication through optical phased array has obvious advantages in energy consumption and control accuracy. Gerchberg-Saxton (GS) is a commonly used modulation phase design method for optical phased arrays. This paper analyzes the relationship between the beam efficiency and accuracy and the number of iterations in the modulation phase of optical phased arrays designed by the GS algorithm. The results show that in the ideal case, the beam efficiency increases with the number of iterations of the GS algorithm, and eventually tends to be stable; the beam pointing accuracy increases with the number of iterations of the GS algorithm, and finally tends to be stable, and the speed of stable convergence is faster than the beam efficiency.

This paper proposes an adaptive error compensation method for wide-range current transformers based on segmentation and dimensionality reduction. On the basis of the circular magnetic split compensation, a segmented aggregation approximation algorithm with a shift factor is used to reduce the dimensionality of the input signal, after which the gain of the compensation signal is further adjusted based on the idea of adaptive control to track the change of the error to achieve closed-loop measurement and extend the range. The validation results show that the error of the low-voltage current transformer for wide-range metering meets the 0.2SS level under both light and full load conditions.

In this paper, we study a distributed resource allocation scheme for unmanned aerial vehicle (UAV) communication systems with parameter uncertainty. A game-theoretic framework is proposed to describe the UAVs’ interactions. Specifically, a distributed robust game is modeled, where the parameters in resource allocation constraints have polyhedral convex uncertainties, which are not exactly known by UAVs. We aim to solve this uncertain problem in the worst case. In this view, we first convert the original robust game into an extended certain game by virtue of the idea in robust optimization. Then we consider distributed dynamics of this certain game to seek generalized Nash equilibrium (GNE) by gradient descent and projected output feedback. Finally, we show several numerical examples to present the feasibility of the proposed dynamics.

In this study, a circuit structure of ring current controlled oscillator (ICO) is proposed. Through the cooperation of counter, bootstrap switch with special timing control and special ICO circuit module, the output frequency accuracy in traditional ring OSC is avoided from being affected by voltage range limitation and comparator delay, the temperature and voltage characteristics of output clock are improved, and the output clock with higher accuracy is obtained under low voltage and low power consumption. The test chip is manufactured by 0.18 µm CMOS process. The power supply voltage is 0.77 V, and its output frequency is 1.32 MHz. The deviation of the output clock with the power supply voltage is within ± 1%, the deviation with the temperature is within ± 2.5%, and the power consumption is 382 nW.

The pattern of exponential number can achieve 99% ideal uniformity distribution level and is characterized with quasirandom nature. In this study exponential pattern is modified to possess an ideal uniform distribution property for operating location permutation. Cipher space of the proposed permutation scheme can achieve its factorial function upper bound. Brute-force restoration of permutation belongs to factorial complexity order which is more complex than exponential function. The proposed encryption scheme belongs to lightweight cryptographic function that is applicable to internet of things (IoT). This novel encryption technique can outperform the advanced encryption standard (AES) from the simplicity of algorithm, the availability, generation and distribution of long encryption key, and high confidentiality level points of view.

In this paper, we study the leader-follower Unmanned Aerial Vehicle (UAV) security game with deception. The robustness under the UAV game with deception describes the model’s ability to maintain players’ profits. We propose a UAV security model with one-leader and multi-followers and define the Deception Strong Stackelberg Equilibrium (DSSE) in the game with deception. Besides, we analyze the robustness of the DSSE to find the boundary that the leader cannot improve its utility from deception.

Signaling System No.7 (SS7) Network was designed with a focus on reliability and efficiency, with a high degree of trust between communication entities and no encryption or defense against attacks. With the development of the network, the vulnerability of SS7 network is increasingly exposed. This paper comprehensively analyzed the security threats existing in SS7 network, and extracted the characteristics of the threats such as location tracking, interception of calls and interception of SMS. On the basis of feature extraction, threat modeling is carried out to extract threat detection rules, and the threat detection model is applied to the experimental network.

Network measurement technology is the basis of the monitoring network devices, link and flow running state and flow behavior model. The current network measurement mainly adopts the method of active measurement, in other words, managers use injection flow probe to measure the network actively, but this method can increase the network flow and influence the state of the network and the network load. To solve this problem, this paper proposes a method that is fine-grained network traffic parameter measurement based on in-band network telemetry technology. Software defined network architecture and the principle of data plane programmable technology is introduced first, and then the process of using in-band network telemetry to obtain network traffic parameters is described in detail. Finally, this paper simulates the network environment and attack environment in the simulation environment to obtain the data set required for the experiment, and the results of the data set are analyzed to verify the effectiveness of this method based on in-band network telemetry technology.

For the current smart high-speed scene, the millimeter-wave radar needs to give the lane number of the vehicle while outputting the lane-level target track information. Since the millimeter-wave radar cannot identify the lane line information, it is difficult to achieve the lane estimation without the help of other sensors. This paper proposes a method to quickly estimate the lane numbers of vehicles in different high-speed scenarios with the help of high-precision maps. First, it unifies the map and vehicle track coordinate system, converts the point set of lane line in WGS-84 format to the Cartesian coordinate system with the radar as the origin after coordinate transformation. Then, through Lagrange interpolation, the X value corresponding to the Y value of the vehicle on the left and right side of the lane line is obtained. Finally, the lane to which the vehicle belongs is estimated by the ratio method. The actual measurement results show that the method can quickly and accurately estimate the lane number of the vehicle in various high-speed scenarios without the help of other sensors.

In Software Defined Network (SDN), the control network is a logically mapped network that is like the brain and control center of the entire network. The reliability of the control network is the ability to survive failures of switching nodes on a large scale. However, there are few studies that consider the control network of SDN as a whole. In this work, we took the SDN control network as a standalone network that is separate from the underlying switching network. We developed a model to evaluate the average reliability of the control network in the presence of multiple failures of the underlying switching networks based on its topology characteristics. Then we searched for the optimal solutions of the model through an opposite learning revised sparrow search algorithm, based on which we proposed a multiple fault recovery oriented deployment strategy of backup controllers in SDN. The simulated experiment shows that the revision of the algorithm is effective and our proposed strategy could help us increase the reliability of the control network to a significant extent.

Secret sharing (SS) technology is widely used in multi-party computation (MPC) because of its simple secret structure and low computational overhead. Aiming at solving the problem that the existing SS-MPC scheme cannot solve the collusion of all computing parties to recover the original data, this work designs a practical SS-MPC for collusion whole (PSCW) scheme. Through the negotiation of redundant computing nodes and redundant security parameters between computing parties, PSCW scheme can ensure that original data cannot be recovered even if all computing nodes collude to obtain all secret shares, which guarantees the confidentiality of the scheme. At the same time, with the addition of redundant computing nodes, PSCW scheme can correctly realize computing requirements of users and ensure the correctness of the calculation results. Compared with the existing SS-MPC scheme, PSCW scheme has higher security and higher application value in real scenes.

As an important link of the Internet of Things, wireless sensor network has great potential in industry, agriculture and daily use of residents. To ensure cloud data security, data is encrypted before being uploaded to the cloud. However, how to ensure the searchability of ciphertext in cloud is challenging. To solve the above problems, a searchable encryption scheme for sensor networks is proposed. The sensor node collects the data, encrypts it and uploads it to the smart contract after collecting the data from the local gateway. In this scheme, the concept of version information is introduced to authenticate ciphertext version so as to realize access control of trap gates of different versions. In order to avoid data leakage caused by the information storage center, the searchable encryption technology is combined with the block chain technology, and the smart contract is used as the data storage center, so as to ensure efficient ciphertext retrieval and eliminate malicious behavior on the server side. Through security analysis, it is proved that the scheme satisfies ciphertext indistinguishability and forward security based on DBDH hard problems. Experimental efficiency analysis shows that the scheme has certain advantages in computational efficiency.

Text classification plays a more and more important role in many practical applications. However, there are a large number of unevenly distributed but important data about the complaints of Chinese telecom users. How to efficiently and accurately implement these telephone consultation contents with only a small number of samples to specific departments and improve the user experience has become an urgent problem to be solved. In this paper, we propose a multi modelbased machine learning method—telecom textCNN predictor which can effectively improve the accuracy and recall of the model on the long tail data, and automatically mark the text, so as to accurately locate the class of the problem. The results show that our model is helpful for business personnel to quickly identify the type of complaint business, conduct targeted business acceptance and user services, and improve user perception and produces better accuracy than the SVM model.

Geographic location privacy protection is an important research content of privacy protection in recent years. Local differential privacy mechanism is one of the mainstream geographic location privacy protection models. among ε-Geoindistinguishability has become the basic research standard of local differential privacy geographic location protection scheme using distance measurement. Combining the ideas of ε-Geo-indistinguishability and Utility-Optimized Local Differential Privacy, we propose a differential privacy concept Utility-Optimized- ε-Geo-indistinguishability which meets the different degrees of privacy protection in different geographical locations. This is the first time in the field of local differential privacy geographic location protection based on distance measurement. At the same time, we propose a localized differential privacy mechanism called Utility-Optimized Planar Laplace Mechanism that can meet UtilityOptimized-𝜀 -Geo-indistinguishability. Theoretical analysis and experiments based on real data sets show that the experimental effect of our proposed mechanism is better than the existing local differential privacy geographic protection mechanism.

With the increasing popularity of unmanned aerial vehicles (UAVs), UAV detection technology has received increasing attention to address security threats. For broad deployment of UAV detection, it is more feasible to perform UAV detection based on existing communication networks than radars. This paper proposes a UAV detection method based on a wireless communication link via a channel state information (CSI) autocorrelation function. The algorithm can be executed in parallel with communication in existing transceivers as no additional signal design requirements and hardware requirements are needed but the CSI, which can be obtained with channel estimation in communication systems. Through experiments, the UAV is successfully detected based on the method proposed in this paper, which verifies its effectiveness.

5G AKA will be subjected to the linkability attacks like Universal Mobile Telecommunications System (UMTS) Authentication and Key Agreement (AKA)because they inherit the authentication failure messages (MAC failure, Synchronisation failure) from UMTS AKA. The existing method usually uses a key to encrypt the failure message to hide the failure type. However, the attacker can analyze the length of the message to distinguish the failure type and then track the target user equipment (UE). In light of this, this paper provides two privacy protection methods to avoid linkability attacks, which can effectively reduce the possibility of linkability attacks. The construction of the two methods and the corresponding implementation steps are analyzed in this paper.

Blockchain technology represented by cryptocurrencies has increasingly become the focus of social attention. The consensus protocol is the foundation for how the blockchain works. PoW, as the most widely used protocol, received more attention from researchers. This paper analyzes the defects of PoW, the most popular public chain consensus protocol in Blockchain, from five perspectives and points out the natural defects of PoW in high energy consumption, electronic waste, carbon footprint, expensive transaction fees, and centralization. This paper encourages the use of PoS and DPoS protocols instead of PoW protocols as they reduce the intensity of competition and may address the root cause of the aforementioned issues.

A quantitative calculation model for the direction similarity of composite spatial objects is introduced in this paper. First, it quantifies the compound expression model, defines the distance between the domain map and direction slices suitable for the model; then, it uses the decomposition idea to decompose the composite spatial object, decompose the reference target, and the source target as a whole; finally, it uses the balanced transmission problem. The minimum element method is used to solve the distance between the direction relationship matrix of the compound expression model, and finally the similarity between the composite spatial objects is obtained. The directional similarity experiment of composite spatial objects at different scales proves the feasibility, and the results are in line with human cognition.

In order to solve the problem that the Topic Talk in the Mandarin proficiency test is still unable to achieve machine evaluation, an intelligent evaluation model for Topic Talk is constructed in this paper. First, according to the requirements of the intelligent evaluation model, a large vocabulary continuous speech recognition module is designed in the front of the model. Then the recognition module is optimized from the two aspects of the acoustic model and the language model. Finally, the experiment is performed on the dataset which is collected from the Mandarin Proficiency Test Center. The experimental results show that the speech recognition module constructed with the improved DBLSTM-HMM acoustic model and n-gram+RNN interpolation language model can better complete the recognition task of front part in the intelligent evaluation model. The improved acoustic model can reduce the word error rate and sentence error rate to 14.08% and 18.22%. The improved language model can increase the word recognition rate by 5.87%, and the correlation between the posterior probability and the manual score by 3.82%.

The South China Botanical Garden is the largest South Subtropical Botanical Garden in China, but it is still in traditional browsing mode. In the botanical garden, there are problems such as confusion of route guidance, single botanical form, and dullness. The rapid development of MR (Mixed Reality) technology brings new opportunities to the tourism industry, and MR technology can bring more interactive and interesting experiences to tourists. Therefore, based on the method of the double diamond model, this study designs a MR navigation system for the South China Botanical Garden to optimize visitors’ experience and improve the service quality of the botanical garden. The research results have reference value for the digital transformation of services in large botanical garden and the improvement of the service quality.

This paper proposes a new model to analysis multi factors that affect internet addiction. Firstly, based on the latest research status and achievements of machine learning, this paper constructs a multi-factor weighted analysis model of Internet addiction tendency. Secondly, in order to analyze the determinants of internet addiction more effectively, this paper designs a deep belief network structure based on the method of machine. Through the training of this network, we can get more accurate weight distribution of various factors in the network structure, so as to extract more effective convolution features of multi-factors. Finally, it gives the compared experimental results which show that this new model can make the internet addiction tendency analysis method be more accurate and effective.

The paper aims to build a traffic prediction model for online car-hailing demand with Improved LSTM and Transformer. There are many factors that affect demand, such as temporal features, spatial features, high signal-to-noise ratio, and so on. In this study, LSTM and Transformer are used to extract the temporal and spatial features of data. The temporal and spatial features are used for bagging ensemble learning to predict the online car-hailing orders. An improved LSTM suitable for Traffic data sets is proposed in this paper. In LSTM, we use wavelet decomposition and reconstruction to extract the small-amplitude high-frequency signal from data. The extracted signal is translated and superimposed into the results of the model. The average MAE of 21.24 is obtained on Online Car-hailing Information Data Sets. Compared with other methods, the results suggest the proposed traffic demand prediction model has better accuracy.

Existing vehicle detection has the problem of unbalanced detection accuracy and speed. Aiming at this problem, this paper proposes a new real-time vehicle detection model named YOLOv3 Tiny Vehicle. The proposed network replaces the Maxpooling layers of the original network with the convolutional layers to ensure that the characteristic information of the vehicle was preserved to the greatest extent. On this basis, our work adds a dense connection structure to the original network, which greatly reduces or even eliminates the overfitting problem during network training. The experimental results show that the mean Average Precision (mAP) of the model on the Beijing Institute of Technology vehicle (BIT-Vehicle) dataset can reach 96.80%, the Frames per second (FPS) can reach 188. At the same time, it also shows that our model has preeminent generalization ability.

Numerical reasoning is a complex subtask in machine reading comprehension, which requires the model to complete discrete operations such as statistics and calculation in the process of reasoning answers. As far as we know, the best solution for numerical reasoning adopts GNN structure. However, the graph structure is often generated by hand-designed rules and it will cost extra computation resource. To tackle these problems, we propose a Transformer-based numerical reasoning solution which transforms the hand-designed graph structure into learnable parameters. Extensive experimental results show that compared to those GNN-based methods, our method can carry out numerical reasoning more efficiently only at the cost of a little decrease on answer quality.

At present, many scholars have done a great amount of research in link prediction. Researchers have found that the attributes of neighbors cannot be ignored in the study of expressing node similarity. Some scholars synthetically consider the degree and H-index to better express node influence. On this basis, the paper introduces a weight factor β to evaluate the role of degree and H-index, then proposes a mixed influence based on diversity of degree and H-index of neighbors model, and carries out experiments on twelve data sets. The experimental results indicate that the link prediction performance can be improved by the weight factor β to the hybrid influence of neighbors.

This paper introduces an objective method to measure students’ achievement. Among many popular methods for evaluating and testing students’ scores, different participants often obtain different results on subjective questions, which obviously lose the objective and fair standard for students’ evaluation. In order to overcome these defects, a new scoring method is recommended in this paper, so that different raters can obtain more objective and fair results in the test of subjective questions as much as possible. In this paper, this scoring method is divided into some main essential steps, which can be summarized as: find out the key points of subjective questions, and divide the questions into several different scoring grades according to different scoring steps.

Aiming at the transaction problem of multi-source heterogeneous distribution network, a P2P intelligent contract based on blockchain technology is proposed. Firstly, a multi-source heterogeneous distribution network model based on Intelligent contract is established. Secondly, a multi-source heterogeneous distribution network market transaction mechanism is proposed. The mechanism designs different kinds of transaction members, frameworks and processes. Transaction members can participate in the market quotation and determine the transaction unit price according to their own power surplus and shortage state. Finally, a smart contract solution with blind beat mechanism is designed for the proposed transaction mechanism. The effectiveness of the transaction mechanism is verified by an example and deployed on the Ethereum private chain. The effectiveness of the transaction mechanism is verified by an example and deployed on the Ethereum private chain. Through the simulation transaction analysis, it can be seen that the above transaction mechanism effectively improves the economic benefits of trading members and effectively ensures the market security and free competition.

Although the lightweight Openpose model can perform real-time human pose estimation on the CPU, it also has the following problems: (1) The ReLU activation function destroys the narrow channel information; (2) Although the model has high inference efficiency, it also has weak feature extraction ability; (3) The model has a small effective receptive field. In order to improve the above problems, we adjusted the structure of the lightweight Openpose model, replacing the backbone of the lightweight Openpose with the backbone of MobileNetV2 to improve the destruction of the narrow channel information by the ReLU activation function. Replacing part of the backbone’s convolutional layers with dynamic depth wise separable large convolutional layers expands the effective receptive field of the model, improves feature extraction and slightly increases the computational cost. Our Lightweight dynamic large convolution model achieves an Average Precision (AP) value of 0.40 on the COCO2017 validation set.

Recently, the transformer-based models have been widely used in sequence-to-sequence (seq2seq) tasks, especially neural machine translation (NMT). In the original transformer, the layer number in encoder is equal to the layer number in decoder. However, the structure is more complex and task is more difficult in decoder than those in encoder, so the layer number should not be same. In order to verify how many layer number in encoder and decoder is properly valued, we improve transformer as our model and conduct four experiments on four translation tasks of IWSLT2017. The experimental results show that the layer number in decoder should be larger than that in encoder, which can bring better translation performance.

Speech recognition has shown excellent performance in recent years, but as we known, this kind of AI task is still supervised and needs labelled data to feed models so that it can achieve a better effect and to inference to get model’s metrics. In actual application scenarios, it always needs many people to check speech recognition’s correctness because of lacking labelled data, it is not realistic to label massive online audio. In addition, this method can’t guarantee comprehensiveness of checking, so confidence estimation algorithm is proposed, which can evaluate speech recognition model’s results and predict error transcriptions automatically. This paper proposed a confidence estimation model based on a multi-feature fusion mechanism and it mainly focus on Chinese end-to-end speech recognition tasks in complex application scenarios. Experiments in Aishell-1 dataset and China telecom’s internal dataset have proved that this model can obtain a good performance.

The skid resistance of the pavement has a significant impact on road safety. To accurately evaluate the change mechanism of skid resistance, a hybrid model based on Bayesian optimization algorithm (BOA) and light gradient boosting machine (LightGBM) is proposed. Firstly, the electric sander is used to obtain the mean texture depth (MTD) of the asphalt pavement. Then, the friction coefficient of road surface under dry and wet conditions is measured by dynamic friction tester at corresponding positions. Finally, a BOA-LightGBM dynamic friction coefficient estimation model is proposed based on the mean texture depth, pavement water content, and vehicle speed. The BOA algorithm is used to adjust the optimal parameters of the model. Compared with the LightGBM model, the results show that the BOA-LightGBM model fits predicted and true values better through the above factors, and the R2 value is 97.21%. It proves that the proposed model has strong stability and nonlinear fitting ability.

Mixup is a learning principle that trains a neural network on convex combinations of pairs of examples and their labels. Despite of its good performance, there is an inherent inconsistency between training and testing in mixup, which makes theoretical understanding difficult and hurts the performance in some cases. In this work, we propose λ-mixup to alleviate this inconsistency. Specifically, λ-mixup reformulates the model to take the interpolation coefficient (𝜆) as input as well, so that a class of models indexed by 𝜆 is learned and we can select one specific coefficient or multiple coefficients for ensembles depending on the testing distribution. We theoretically demonstrate that, with enough data and model capacity, λ-mixup can recover the original conditional distribution. Moreover, we conduct image classification tasks on multiple datasets, including CIFAR-10, CIFAR-100 and Tiny-Imagenet, showing that comparing with mixup, λ-mixup exhibits better generalization, calibration and robustness to adversarial attacks and out-of-distribution transformations.

In order to improve the firing efficiency of large-caliber naval gun, a new method of “Time-on-target (TOT)” for fixed-loading ammunition of large-caliber naval gun is proposed in this paper, and a model of TOT for fixed-loading ammunition of largecaliber naval gun is established, which verifies the feasibility of multiple simultaneous firing. Based on the naval gun attacking small and fast targets on the sea surface, combined with the naval gun tactical application, the TOT firing method of fixed ammunition of naval gun based on ladder firing is given, and the effectiveness of the method is verified by simulation.

Despite the impressive improvements of Visual Question Answer (VQA), it still remains a challenge of how to avoid the suffering of spurious correlations from textual content to answer. Previous researches have shown that due to the existence of language bias in the VQA dataset, VQA models may tend to capture superficial statistical correlation and suffer from the poor generalization capability in the out-of-distribution data. To alleviate the biases caused by language modality, we propose a method of context augmentation and adaptive loss adjustment, which can alleviate shortcut learning behavior of VQA models. Specifically, the existence of language bias is due to the high co-occurrence frequency of categories and the words in “Question”, therefore, we propose to use “Paraphrase Generation” to produce paraphrases with diverse contexts, so as to mitigate such correlation. Secondly, we use adaptive loss adjustment to adjust the importance of samples, that is, reduce the importance of bias-aligned samples and improve the importance of bias-conflicting samples, so as to guide the model to capture the intrinsic attributes that are beneficial to generalization. The experiments have demonstrated the feasibility and validity of our method on a variety of VQA models.

Pythagorean fuzzy set (PFS) is applied to the problems of multi-attribute decision-making, and a similarity measure of grey relational analysis (GRA) and PFS based novel method for MADM is presented, and the corresponding algorithm is designed. Firstly, a new similarity measurement method is proposed, and some important properties are proved. Furthermore, the ranking of decision options is achieved by comparing the grey relational similarity of each option with the optimal and worst ideal sequences of the PFS when the attribute values are considered as the Pythagorean fuzzy number (PFN). Finally, a detailed example and comparative experiments are illustrated to prove the effectiveness and correctness of the proposed method.

To protect coastal areas and ecology environment, Sand nourishment is widely used as an effective method. The morphology evolution and prediction is needed for sand nourishment and maintenance. Based on the observations of a mega-nourishment, a data-driven approach combining temporal and spatial features for the prediction of nourishment morphology evolution is proposed in this paper. The sequence grids from the survey data are used to extract the spatial features of nourishment beach by CNN, which can reduce the complexity of the network and extract the spatial features. And then LSTM is employed to derive the temporal relationship from past time-series features for predicting future nourishment terrain. Finally, the final prediction result is obtained through decoding the LSTM output by the fully connected layer. The complex spatiotemporal correlations of the sand nourishment evolution can be derived by training the proposed model effectively. The proposed model achieved the best performance for nourishment morphology evolution prediction compared with other methods as shown by the experimental results.

The renovation of existing residential district is being promoted in major cities across China, including Shanghai. Therefore, it is important to improve the thermal environment in surrounding areas of buildings. Sensible heat flux from building external surfaces is a key factor influences the thermal environment. In order to clarify sensible heat flux, it is significant to understand the distributions of Convective Heat Transfer Coefficient (CHTC) and surface temperature. This study used 3D CAD-based thermal-environment simulator which can calculate building surface temperatures by heat balance simulation. By coupling the heat balance simulation with Computational Fluid Dynamics (CFD), the study examines the relationships between wall-adjacent wind velocity and CHTC to derive a prediction equation which can apply to the prediction of CHTC for High Reynolds number k-ε model (the High-Re model). Finally, the study examines the prediction accuracy of the prediction equation by comparison with the Jürges equation. The results show that the prediction equation for the High-Re model is appropriate for predicting the sensible heat flux from the building external surfaces under the forced convection field.

Modular shipbuilding can effectively shorten the cycle of construction and improve quality of the product, however stricter requirements of the accuracy control for product have been proposed at the same time. Taking a marine power shafting as an example, the influences of part manufacturing, positioning and matching errors on the axial and radial deviations of the stern and power shaft were studied based on 3DCS. The main factors affecting the assembly accuracy and their contribution were determined. It was found that the sealing device was the most influential factor for the deviation of stern and power shaft. The contribution of sealing device to the radial and axial deviation of stern shaft were 99.45% and 83.12% respectively, and to the radial and axial deviation of power shaft were 98.91% and 74.21% respectively. The research results would provide guidance for accuracy control of marine shafting modular construction.

The intermediate joint is the weakest part of the insulation in the cable line. How to efficiently and reliably transmit the status information of the cable intermediate joint is the key in the online detection of the cable intermediate joint. By analyzing the characteristics of power line channel and the coupling principle of inductive coupler, this paper builds the power line communication (PLC) system model between cable intermediate joints based on orthogonal frequency division multiplexing (OFDM), considers the influence of inductive couplers at the transmitting end and receiving end, and simulates and analyzes the relationship between different signal-to-noise ratio (SNR) and bit error rate (BER) curve. The simulation results show that the PLC system with coupler can reduce the BER and improve the transmission performance of the system by changing the encoding method and modulation method, which provides a new idea for the effective transmission of online detection data of cable intermediate joint.

As a non-negligible problem, the suppression of sloshing is very important for liquid storage carriers such as LNG (Liquefied Natural Gas) ships. The effective suppression of impact pressure caused by sloshing liquid improves the safety of the structure. In this paper, the porosity characteristic is combined with the adjustable hanging part, and the effects of this composite anti-sloshing method on sloshing states are studied by CFD (Computational Fluid Dynamics). The results show that, compared with the traditional solid part, the composite method with appropriate porosity can further suppress the sloshing, and the load acting on the hanging porous part induced by the sloshing liquid is smaller. When the porous part near the static free surface moves downward, the damping effect enhances, leading to further improvement of the anti-sloshing effect, but the load acting on the part increases.

Optical scattering caused by the high-speed turbulence will reduce the accuracy of optical sensors on aircraft. The existing research mainly focuses on the scattering field on the macro scale. A new method to analyse the photon-scattering characteristics in high-speed turbulence from the micro point of view is proposed in this paper, which can explain the micro disturbance of airflow molecules to photons. In order to construct the relationship between micro disturbance and macro distortion, molecular clusters are proposed to describe the multi-molecular scattering field, and the scattering characteristics of molecular clusters with wavelength, density and vortex shape are obtained by statistical method. The simulation results show that the coherent superposition of scattered waves is the dominant factor of the micro scattering characteristics of molecular clusters. The method proposed in this paper provides a new idea for the study of aero-optics under complex flow conditions.

Although the rapid development of information technology makes rooms for many small and medium-sized enterprises on program deployment through Virtual Machines (VMs), the physical resources is not fully occupied since some programs are not as large as the virtual systems. Therefore, we applied the Docker container technology, which is an application-level isolation. It is more flexible and smaller than virtual machines. This paper studies the standardized migration, automatic deployment and application cluster monitoring of Docker container technology and applies them to the Saas platform. It provides visual operation function for enterprise project management and reduces the steps of system deployment, operation and maintenance.

With the relatively uneven distribution of medical resources in China and a new outbreak of COVID-19 at the end of 2019, we developed an intelligent medicine cabinet to alleviate the problem of high pressure and difficulty in accessing medical care in hospitals around the country. The medicine cabinet has a signal transmission circuit system based on 51 microcontroller and a new X-Y trajectory control module, which controls the Pulse Width Modulation (PWM) signal by Proportion Integration Differentiation (PID) algorithm to improve the accuracy of the DC motor. It has the functions of online drug selection, drug sales, drug transmission, etc. Meanwhile, the online drug purchase system based on the WeChat applet can reduce the probability of infection by a new coronavirus. And the new X-Y cargo track will significantly improve the safety of fragile drugs while ensuring their delivery. The development of this medicine cabinet will greatly reduce the operating cost of pharmacies and meet the demand of people to purchase medicine at night.

Radar signal recognition is a key step of electronic reconnaissance. In order to use various parameters of radar signal to identify its country in complex electromagnetic environment, this paper analyses the characteristics of radar signal and the development status of related recognition methods. Aiming at the support vector machine (SVM) classification algorithm in machine learning, this paper studies the feature selection and the establishment, evaluation and optimization of model, and completes the programming and application of the algorithm, the recognition accuracy and efficiency are improved. The research shows that when SVM is applied to radar signal recognition, the classification accuracy is more than 93% and the AUC value is more than 0.99.

The harmonic oscillator is an important and typical physical model in quantum mechanics and quantum optics. It is very important and widely used and has been confirmed by the development and application of science and technology. The harmonic oscillator in the elliptic paraboloid potential is studied and effects of the elliptic oblateness on the harmonic oscillator in the elliptic paraboloid potential are revealed. The energy of the harmonic oscillator in the elliptical paraboloid potential is quantized, which is described by two quantum numbers n, and m. Generally speaking, the maximum value of the probability density peak decreases as the extremum number of the probability density increases. However, this reduction is with oscillations and fluctuations, which shows even a maximum structure for the smaller quantum number. For the elliptic paraboloid potential, the spatial distribution of the probability density on different cutting surfaces is various. The flatter the ellipse is, the greater the probability density of the ellipse center, and the smaller the extreme of the edge peak of the probability density will be.

An increasing number of various sensors in metallic environment such as vehicle, spacecraft and industrial equipment has contributed to the application of capacitive coupling communication technology, which is considered to be more useful compared with wired communication or traditional wireless communication. In this study, the composition and structure of the communication system were proposed firstly. Then the channel transmission characteristics with different electrode parameters were obtained using finite element simulation model. Moreover, vector network analyzer was used to test the path loss for verifying the simulation findings. The experimental results are consistent with the simulation results, and showed that the transmission gain of the system varied with the electrode parameters. This study can provide theoretical foundation and experimental support for capacitive coupling communication in a metallic environment, and has great significance for promoting the transceiver design and system optimization in the future research.

Many malware families could generate a huge number of pseudo-random domain names through DGAs (Domain Generation Algorithm). Using DGA domain name to take DDoS (Distributed Denial of Service) attacks makes network defenses more difficult. So detection of DGA domain name has become an important research in network security, and methods based on neural network have been explored. By extracting different character features of domain name in character-level word embedding, this paper compared the performance between CNN (Convolutional Neural Network) and Bi-LSTM (Bi-Directional Long Short-Term Memory) in two-classification of DGA domain name. Experiment results show that using character features including semantic features could improve the performance of neural network, and there is little difference between CNN and Bi-LSTM in DGA domain name detection.

This paper proposes a time-sensitive network performance test program based on a real-time edge computing platform. It uses a real-time data engine with a time-series database as the core to realize the collection and status monitoring of the real-time communication flow of the time-sensitive network. The clock synchronization is carried out by building a specific test experimental platform and end-to-end latency performance testing, and the results show that TSN network can ensure the delay of the time sensitive flow in the transmission process and the real-time arrival of the message in the complex network environment.

This paper proposes a study on whether the speaker’s body size (height, weight) and oral cavity (lip protrusion LP, lip opening LO, front cavity FC) characteristics can be predicted based on the acoustic features of speech. Firstly, Pearson’s correlation analysis was first conducted to examine the relationships between acoustic features and body size and oral cavity characteristics. Further, the effects of acoustic features in predicting body size and oral cavity characteristics were examined using random forest and decision tree models. The results showed that fundamental frequency statistics (i.e., mean, max, min) exhibited significant negative correlations with height, weight, and FC. Besides, good accuracies of classification in height, LP range, LO range, and FC range could be achieved based on the acoustic features. The findings in the current paper imply that acoustic features could be the potential features for identification of the speaker’s body size and oral cavity characteristics. This paper will not only contribute to the research and practices in forensic speaker profiling and but also provides foundations for the technology of automatic speaker recognition.

In the paper, we establish a seepage model which satisfies a constant rate, the rate is generally called outer boundary conditions including the infinite condition, the closed condition and the constant condition, and also the rate excludes the effects of the skin and wellbore storage effects. This model can help us to solve the double permeability reservoirs problem under the condition of the plane radial flow. Firstly, this model produces non dimensional treatment by introducing non dimensional quantity, and then obtains a problem which has boundary value in Laplace space; secondly, we can use a method of Laplace transform in this model; thirdly, the solutions we get are proved to have a similar structural equation. Therefore, a new method for solving such seepage model is obtained-similar constructive method. And also, this paper transforms the problem into the problem which has boundary value and can avoid the complicating calculate of the model. As an innovative approach, similar constructive method can be obtained easily to construct with a simple arithmetic, to be mastered and be used in many practical problems in engineering optimization.

COVID-19 has caused a large number of online public opinion incidents. How to timely and effectively guide the resulting network public opinion has become an urgent problem to be solved. This paper collects more than 130,000 original Weibo posts during the Wuhan “city closure” incident, and analyses the topic characteristics of the incident on the basis of user classification through topic models and community detection algorithms. It was found that during this period, the government responded quickly to the epidemic and gained public support. For different Weibo users, officially certified users mainly publish information about the epidemic and epidemic prevention measures. Personally certified users mainly forwarded and transmitted official information actively, and they also expressed their opinions and made suggestions. Non-certified users actively expressed their emotions and opinions, so they were important users that reflect public opinion.

Multi Variant eXecution (MVX) is a security defense technique that uses software diversity to protect system from attacks. MVX improves security capability by enhancing system endogenous security compared to traditional passive defense techniques. However, the current MVX technique lacks formal theoretical analysis and cannot effectively assess the overall security of the system. To address the constraint relationship between complex attack means and dynamic defense environment, we construct a novel atomic combination attack chain model, which decomposes macro attack means into single atomic attack behaviors and provides theoretical support for analyzing the security capability of dynamic systems. Then, the defense model of the MVX system is established, and the defense model’s security capability is analyzed using the attack model. Finally, the advantages and shortcomings of the security defense capability of the MVX system are evaluated based on a typical kernel attack example, and system optimization improvement measures are proposed.

To enhance the precision and efficiency of manipulation trajectory tracking, a trajectory tracking control strategy combining nonlinear disturbance observer and adaptive inverse sliding mode control is designed. Prior to designing the operator trajectory tracking control strategy, the general kinetic model of the manipulator must be constructed. For the observable part of the disturbance signal, the nonlinear disturbance observer is used for online observation, and the observation error converges exponentially by selecting appropriate parameters; For the unobservable part, the adaptive law is used to compensate, and the Lyapunov stability theory establishes the system’s stability. According to the simulation results, the other two methods, the approach suggested in this paper not only speeds up the response speed of the system but also can better overcome the uncertainty of the object and external random interference, significantly reducing system chatter and enhancing the system’s ability to exert control.

The increase in greenhouse gases has an impact on climate change, and forests play a significant role in the process of carbon sequestration. To quantify the effectiveness of forests in carbon sequestration, this paper constructs carbon sequestration models for forests and is evaluated using the Daxing’an mountains. We randomly select 57 countries, and a comprehensive evaluation of the ecological and cultural value of forests is conducted using the Topsis model combined with the entropy weight method. The results show that the model can maximise the combined benefits of forests in terms of carbon sequestration, culture and ecology, and is conducive to the sustainable development of society.

Software can be improved through user experience metrics, so it is very important to choose the right measurement method. Through the desktop research and other research methods, the user experience research methods and model results are summarized. With software A and B as the research object, according to the product form and the characteristics of the business, the user experience evaluation model commonly used in the industry is used to measure the two software products by means of usability test and questionnaire data from the various attributes and indicators of the evaluation model, so as to the rationality and feasibility of the evaluation model and provide Suggestions for improvement as the research object.

In the 21st century, human beings have entered the information age. Knowledge and technology have become the core competitiveness. Computers are widely used in all walks of life and become an indispensable thing in our lives. People's thoughts, inner spiritual needs and artistic aesthetics are also imprinted on it. The imprint of the information age. In recent years, 3D printing technology has penetrated into our lives, and artists are constantly experimenting due to its novel molding methods. The main purpose of this paper is to study the application of computer 3D printing technology in jewelry processing models. Starting from the advantages of computer 3D printing technology, it analyzes how computer 3D printing technology is applied in the processing model of jewelry, and the necessity of application, through the comparison of casting model statistics and casting methods.

In this research, machine learning algorithms such as decision tree, random forest, and BP neural network are used to predict a certain dataset, and then a voting prediction model is built based on the above three machine learning algorithms. To verify the performance of this voting model, we introduced confusion matrix and F1 score to evaluate the effectiveness of machine learning. The experimental results show that the performance of the machine learning strategy based on the voting model outperforms that of a single machine learning algorithm and that adjusting the voting weights of a single algorithm can also affect the performance of the whole model. This result is well worth further study.

This research used the machine learning algorithm of the BP neural network to predict a data set. In order to verify the performance of the prediction model, we introduce the confusion matrix and F1 score to evaluate the effect of machine learning. In order to optimize the BP neural network model, we use feature engineering to process the data set and apply the BP neural network model to this new data set. The experimental results show that the machine learning performance of the BP neural network model based on feature engineering is improved.

In complex environments, UAV wireless sensing networks suffer from path loss, incomplete channels, swarm access security and other problems, and the signal fading and packet loss rates are very obvious. In order to solve the problem of user control data security, this paper gives full play to the characteristics of blockchain technology and builds a blockchain-based UAV cluster anti-interference communication network in UAV cluster applications to achieve antiinterference performance. Meanwhile, Markov decision process is used to optimize and realize the research of UAV airground integrated radio wave anti-interference.

Aiming at the problems of low control accuracy and low execution efficiency of air defense anti-missile concentrated fire shooting strategy, this paper studies the process mapping and execution method of air defense anti-missile concentrated fire based on the combination of OODA loop and BPMN. Firstly, the concentrated fire shooting model is established, and BPMN model of concentrated fire shooting process is built according to OODA loop Theory, then the process model is executed by Cloud Native Command and Control platform, and the execution efficiency and control precision of the model are analyzed. Compared with the traditional concentrated fire shooting scheme, the concentrated fire shooting process modeling with OODA-BPMN and Cloud Native theory has the characteristics of module replaceable, high control precision, fast response and multi-instance parallel execution.

Scanning Acoustic Microscopy (SAM) is an essential tool for the non-destructively deriving depth-specific information, which can detect the internal morphology and defect location of Integrated Circuit (IC) sensitively. When SAM equipment works, it will generate a large number of acoustic scanning images, which provide data support for defect detection algorithm. However, due to lack of professional researchers to uniformly sort and classify these acoustic scanning data and the deficiency of standard and available acoustic scanning image datasets, it is impossible to carry out the research on intelligent detection algorithm. In order to solve the above problems, a novel Multi-category Scanning Acoustic Image (MSAI) database is presented in this paper. MSAI database including 4565 acoustic scanning images acquired from 52 products in four packaging structures (BGA, QFN, SOIC and SOP). In order to prove the availability of MSAI database, four typical convolutional neural networks are used to identify and classify the MSAI database. The experimental results show that the VGG-16 model achieves the best classification performance in IC packaging structures grading, which show a train accuracy of 99.58% and a test accuracy of 99.44%, and all network models show a good inter-class separability on MSAI database.

The aim of this study is to research the features of white matter in the brain by Diffusion Tensor Image (DTI) from the patients with mild cognitive impairment (MCI) and using the features to identify MCI and Normal Control (NC) to explore new methods for MCI diagnosis. In this study, 38 brain DTI images of MCI patients and NC were extracted respectively, and the parameters of cerebral white matter fiber tracts were analyzed. Using automatic fiber tract quantification (AFQ) technology, the index values with significant differences between MCI patients and NC were calculated. Support Vector Machine (SVM) model was built to classify MCI patients from NC. We found significant differences in right cortical spinal tract (CST_R), right uncinate fasciculus (UNC_R), left internal fronto-occipital tract (IFOF_L), and Callosum Forceps major (FP) in MCI patients and NC. The classification accuracy, sensitivity and specificity of the training set and test set were 94.73%, 92.11% and 97.36%, respectively. This study demonstrates that there are significant differences in certain fiber tracts in MCI patients compared with NC and using these fiber tract groups can effectively classify the MCI patients and NC, which can provide novel information for MCI white matter decline.

Human temperature characteristics are closely related to physiological activities such as respiration and heart rate. For daily healthcare needs, a real-time contactless monitoring of respiration rate and heart rate based on facial thermal imaging was proposed. A RetinaFace framework was applied to detect the facial region from the video frames of thermal imaging. The facial region was divided equally into 12 areas. For the detection of respiratory rate, the respiratory signal is obtained by detecting the nose area and the feature tracking recognition method based on the optical flow field method. For the heart rate detection, an independent component analysis (ICA) was performed on the 12 areas. The heart rate was calculated based on the region with the higher proportion of energy according to the ICA. Experimental results showed that real-time contactless monitoring system proposed in this study can accurately detect the respiratory rate and heart rate, which may facilitate the daily healthcare examination for a large population.

With the continuous development of remote sensing technology, remote sensing images show multi-source, massive and high resolution, resulting in exponential growth of remote sensing data, which brings great challenges to image pyramid storage and retrieval. In order to solve the problem of low storage and query efficiency of tile data, the Z-Curve index is proposed in this paper by improving the Z curve. This paper uses the distributed Accumulo database for storage. The experiments show that this method can effectively improve the efficiency of tile data query.

Improving the invisibility and robustness of watermarking, increasing the embedding capacity of watermarking and reducing the complexity of watermarking algorithms are becoming hot topics and difficulties in the research of watermarking algorithms nowadays. In this paper, we propose a digital watermarking technique based on discrete wavelet transform for the protection of author’s copyright. First, the watermark is pre-processed, specifically, then the Arnold transform is used for encryption to ensure the security of watermark. Then the wavelet coefficients are obtained by discrete wavelet transform of the original image. Finally, the watermark information is embedded into the wavelet coefficients of the image. The watermark embedding experiments indicate that the algorithm is reversible and the invisibility of watermark is excellent. The watermark extraction experiments demonstrate that our method has strong robustness to against various external attacks.

Today, medical and health care is playing an increasingly important role in people’s lives, the number of medical images increases rapidly, and the traditional manual diagnosis can’t meet the increasing demand of clinical diagnosis. It is the core of this research to realize the localization and classification of lesions by training a neural network, and to assist doctors in clinical diagnosis. Based on the one-stage object detection network FCOS, this paper improves the detection accuracy to a level comparable to the two-stage detection network while taking into account the detection speed and achieves a mAP of 0.664 on the MRI test set, which is higher than most existing networks. Overall, the proposed network has good performance in clinical assistance, and can be competent for some clinical application tasks of real-time detection.

The development of target detection and recognition algorithms in the field of image processing has promoted the development of automatic image conversion systems for digital musical scores and related algorithms. This was crucial for the development of note-recognition coding and the conversion of musical scores into readable digital formats. The purpose of this paper is to study the automatic conversion system of digital music score image based on image recognition algorithm. The deep learning technology based on computer vision is used for musical score image recognition, and a digital musical score image automatic conversion system is designed and built. Harmonic peak method and confidence method are introduced, and their shortcomings are analyzed. Combined with the advantages of these two algorithms, an improved normal harmonic method is proposed, and the detailed ideas and steps of the algorithm are given. And the extracted fundamental frequency and time information are converted into pitch and time value combined with the basic knowledge of music theory. The recognition accuracy rate of the normal harmonic method proposed in this paper reaches more than 92%, which not only greatly improves the recognition accuracy, but also retains the strong anti-noise characteristics of the traditional harmonic peak method.

In order to improve the overall accuracy (OA) of the AlexNet model for high-resolution remote sensing scene images with complex backgrounds, we proposed an improved remote sensing scene image classification model. Firstly, we used Layer Normalization (LN) to replace the Local Response Normalization (LRN) in AlexNet and changed the convolution kernel of the first convolution layer to 7 × 7. Secondly, to focus on critical information in the feature extraction process, and suppress irrelevant background information, the two attention modules of Convolution Block Attention Module (CBAM) and Squeeze and Excitation Module (SEM) were combined. In this study, the classification verification was performed on three remote sensing scene datasets of NWPU-RESISC45, AID, and UCM, and achieved 96.29%, 96.02%, and 96.57% overall accuracy, respectively. Compared with AlexNet, the OA improved by 14.38%, 12.09%, and 9.9%, respectively, therefore, the improved model of this study can significantly distinguish between object information and background information in remote sensing scene imagery.

In recent years, with the booming development of deep learning, content using images has become a mainstream research hotspot for retrieval. Fabric image retrieval, as a specific application scenario for retrieval, has potential value in many areas such as textile product design, e-commerce and inventory management. However, in fabric retrieval, fabric image has unique striped texture features, and it is still a challenging task to better extract its features for retrieval. To address these problems, this paper proposes a method for fabric retrieval based on a dual attention mechanism. The method inserts a dual attention mechanism module into the convolutional neural network, firstly strip attention module, which connects long and distant image information to capture narrow feature regions, and secondly the channel attention module, which adaptively recalibrates the channel feature responses and selects the prominent and effective feature maps. Experimental results show that the proposed method performs superiorly in fabric image retrieval, and DAMNet is proposed to solve the problem of feature extraction and retrieval of such fabric images.

Human action video recognition focuses on discussing the theories and new algorithms to make computers understand the meanings of variety videos, such as those from TV, sports or family videos, and recently it plays an important role in advanced human-machine interface, security monitor, video retrieval and so on. The step-by-step process of traditional computational algorithm for human action recognition from videos are described in this paper, and some important deep learning models are also addressed by which some advantages and disadvantages of each approach in different tasks are analysed. Finally, some current trends and prospects of video recognition models are concluded.

Observing the structure of retinal blood vessels can help doctors diagnose the disease of patients, so the accurate segmentation of retinal blood vessels has important research significance. However, there are many problems in retinal vessel segmentation, such as complex and small vascular structure and low image contrast, which lead to low segmentation accuracy. To solve the above problems, this paper proposes an attention-directed adversarial network for retinal vascular segmentation. The purpose is to guide the network to learn useful information for vascular segmentation and ignore useless redundant information. The attention directed generative adversarial networks consist of generator and discriminator. The generator uses U-Net architecture and combines the high-low feature attention modules. The high-low level feature attention modules act on the high-level and low-level feature maps so that the model can strengthen the high-level and low-level spatial features respectively, eliminate redundant information, and guide the model to pay more attention to the vascular foreground information. The batch normalization layer in the generator is also removed to avoid the impact of unstable batch statistics on the segmentation results when the generator is trained in small batches. The discriminator consists of a stack of residual modules, which together with the generator form a conditional generation adversarial network. The experimental results show that the Se, Sp, Acc and AUC of this paper’s method tested on the DRIVE dataset are 82.88%, 97.45%, 95.59%, and 97.86%, respectively, and all the indexes are better than the current mainstream retinal vessel segmentation algorithms.

Aiming at the disadvantage of poor stability of universal measurement matrix, this paper proposes an optimization method of measurement matrix, which is called Schmidt orthogonalization method. Row vectors are orthogonalized step by step to reduce the correlation between column vectors of measurement matrix. The simulation results show that under the same sampling rate, the one-dimensional signal reconstructed based on the optimized measurement matrix has higher reconstruction accuracy and quality, and the number of measurements required for accurate reconstruction is also reduced; the two-dimensional image reconstructed based on the optimized measurement matrix has higher signal-to-noise ratio.

With the continuous improvement of teaching contents and methods, the concept of students as the main body of teaching has been deepened. Efficient and accurate identification of students’ classroom concentration has become an effective means to improve classroom efficiency and is also a focus of attention in the field of education. In this paper, we propose a concentration recognition model based on the multimodal fusion of students’ head posture and facial expressions and validate and experiment on a self-built concentration database. The experimental results demonstrate that non-negative emotions are beneficial to concentration, and the student’s head orientation can reflect the student’s concentration more clearly. The multimodal fusion method performs better than the unimodal method. This will reduce the workload of manual data collection and effectively improve the accuracy of data collection. It also strengthens the home-school association and has implications for the development of educational theory and practice.

Artificial intelligence (AI) technologies have been used to develop and advance healthcare industry. Typical application scenarios in medical sciences include diagnosing patients, end-to-end drug discovery and development, improving communication between physician and patient, transcribing medical records, and remotely treating patients. This paper aims at introducing AI applications in the field of medical sciences, especially the areas of pharmaceutical and medical imaging. It first introduces the current AI technologies used in medical science in general, and then showcases these applications in pharmaceutical and medical imaging areas. Insilico Medicine (Hong Kong) and Pfizer (US) are the illustration cases from pharmaceutical industry; and Deepwise (China) and GE healthcare (US) are selected cases from medical imaging industry. The comparison within each industry and between two countries (China and US) are summarized at the end.

Fluorescence molecular tomography (FMT) is a powerful modality for resolving the three-dimensional (3D) distribution of fluorescent targets inside biological tissues. However, the inverse problem of the FMT is severely ill-posed due to the strong scattering effects of photons inside biological tissues. Previously, regularization-based methods have been widely used to mitigate the ill-posedness of FMT. Due to the complex iterative computation and time-consuming reconstruction process, the FMT remains an intractable challenge for achieving accurate and fast 3D reconstructions. In this work, we propose a multi-attention prior based residual encoder-decoder network (MAP-REDN) to perform FMT reconstruction. Firstly, the multi-attention mechanism can provide weighted a priori information to the fluorescence source, enabling MAP-REDN to effectively mitigate the ill-posedness and enhance the reconstruction accuracy. Secondly, since the direct reconstruction strategy is adopted, the complex iterative computation process in the traditional regularization-based algorithms can be avoided, thus tremendously accelerating the reconstruction process. The experimental results demonstrate the feasibility of the MAP-REDN in achieving accurate and fast FMT reconstruction.

In this paper, we propose an efficient framework to estimate the motion of a robot when it is navigating indoors. Particularly, we mainly focus on the case where the robot is mounted with an RGB-D camera, and the environments have a lot of planar structures such as the floor, ceiling and walls. The novelties of our method lie in three-folds. First, an efficient normal vector extraction method for planes is proposed to fully make use of the planar structures. Without fitting planes, we only need to use the inverse-depth induced histograms to get the dominant planar structures as well as the normal of the planes. Second, since the robot is roaming indoors, we assume that the environment satisfies the weak Manhattan world constraint, i.e., the floor and ceiling are perpendicular to the walls. It is reasonable in our real world. Based on this assumption, we can calculate the relative camera motion by aligning the current local frame with respect to the world coordinate. Third, we present real-world RGB-D datasets that satisfy the weak Manhattan world constraint including 5,737 images, which make contributions to the community. Extensive experimental results show a very promising performance of our method in terms of accuracy, robustness and efficiency, especially in large-scale lowtextured scenes.

The completion and detail retention of face replacement after face changing are difficult technical problems in face swapping. This paper proposes an efficient and flexible model for face swapping which can realize high-quality face swap and improve the quality of the outputs of the existing networks. The proposed face swapping network is completed by the AEI-Net. After training the AEI-Net (the AEI-Net consists of three subnets: identity encoder, multi-level attribute encoder and ADD generator), the generated image with face change effect can be obtained. In our experiment, we use the training data sets which are CelebA and FFHQ as other face swapping networks to train their model and our model and use the same test data set to evaluate it. After training, it can recover the abnormal area in a self-supervised way. The results show that compared with the state-of-the-arts, our model achieves good performance in terms of the realism of the generated face and the degree of detail reduction.

Recently, unmanned aerial vehicles (UAVs) have been used to conduct the task of power line inspection. Unfortunately, most of the images acquired by UAVs are invalid for visual inspection. The main reason is that the collected images are low quality and repetitive. To obtain a set of valid images with high quality, a novel multi-information fusion perception (MIFP) model is proposed to automatically clean the large-scale aerial image data. Firstly, the image quality features and image content features are extracted, in which the weights of different features are used to evaluate the effect on the final image quality. Secondly, the image spatial features are exploited to aggregate spatial information in weight maps of different sizes. Then, the image quality and content features are merged into the multi-information features that characterize the quality of the final image. As a result, the image is picked out according to the quality score. Finally, experimental results show that the proposed multi information fusion perception model has excellent performance on real databases.

Bronze Inscriptions is essential to the study of ancient Chinese history. We have obtained a set of materials containing 102,338 images of bronze texts collected by experts, but the distribution of these acquired images is imbalance, with a large sample of common words but the opposite for rare words. To solve this problem, we have adopted an end-to-end enhanced rare data generation module that can generate feature maps for rare classes during the training phase. The experiment results show that the proposed model can obtain an overall average accuracy of 77.38% and a minority accuracy of 68.24%.

This paper analyzes several commonly used extraction methods of surface water bodies information, including the CIWI (Combined Index of NDVI and NIR for Water Body Identification) model, normalized difference water index (NDWI) and near-infrared band model and the multiband spectral relationship method, and compares the effects on water body identification in the boundary river basin based on the accuracy results. The results show that the new type of water index—CIWI model has advantages in the water-body information extraction. This model can effectively separate the information of waterbody, vegetation, cloud shadows, urban land and other information, with extraction accuracy more than 90%, achieving the extraction of water-body information in the study area.

Using autoencoders to reconstruct current video frames or predict future frames is a popular method in the task of video anomalous behavior detection based on weakly supervised learning. Recent studies have shown that introducing a memory module into an autoencoder can capture a limited number of normal patterns and cannot cope with new scenarios in the test set. Therefore, this paper proposes a dual-stream based multi-level memory-enhanced conditional variational autoencoder model (TS-MemCVAE), which uses RGB image and optical flow image dual-stream input, and adds memory modules at the bottle-neck. The memory module contains normal mode features of different sizes. At the same time, with the aid of optical flow information, the model can sensitively identify abnormal behaviors with large reconstruction errors. The model is divided into two parts: a multi-level memory-enhanced auto-encoder and a conditional variational autoencoder. The former is responsible for the reconstruction of the input video, and the latter is used to capture the high correlation between the reconstructed video and the optical flow image. further predictions. The model is validated on two benchmark datasets, UCSD Ped2 and CUHK Avenue, and achieves 95.83% and 84.16% on AUC, respectively, and its excellent performance proves the effectiveness of the model.

Gastric cancer is the fifth most common cancer and the fourth highest cause of cancer death in the world. One molecular subtype of gastric cancer called Epstein-Barr virus (EBV) positive tumor often responds remarkably well to immune checkpoint inhibitors and has a favorable prognosis. Considering EBV testing is often time-consuming and costly, it is of great significance to develop an automatic classification method for EBV subtype prediction based on only the cost efficient pathological images. In this study, a self-supervised learning method was proposed to train a more generalizable feature extractor (often consisting of multiple convolutional layers) using only unlabeled pathological images, based on which the classifier head (often consisting of two- or three-layer fully connected layers) can then be more efficiently trained using a large number of labelled pathological image patches. In particular, a novel formation of positive pairs for self-supervised learning was proposed by considering that neighboring patches in each pathological image often share similar visual features and therefore should have similar feature representation from output of the feature extractor. In addition, by imitating the diagnosis process of pathologists who often observe the pathological image at multiple magnifications, a multi-scale ensemble model was proposed, with each individual classifier for prediction of image patches with a unique magnification scale. Experiments on two external pathological image datasets show that the proposed self-supervised learning can help gain a more effective EBV classifier and the multi-scale ensemble model can further improve the prediction stability.

In vehicle detection and tracking methods based on roadside cameras, due to factors such as motion blur, vehicle occlusion, and target scale changes in the video, problems such as missed detection, false detection, and low positioning accuracy may occur during vehicle detection and tracking. This paper proposes to use the improved target detection model YOLOv5s as the target detector, then combined with the classical Deep SORT target tracking method, to achieve end-to-end vehicle detection and tracking. By integrating the attention mechanism with the detection network, and modifying the loss function, the ability of the model to extract features is strengthened, and the final vehicle detection accuracy is 96.7%. During tracking, the vehicle IDs are reduced to 28 times and the operating speed reaches 32 Hz.

As a structured abstraction method for objects and their interactions in visual scene, scene graph captures entities in the scene and the relationships between the entity pairs, and helps in understanding the visual scene better. Currently, scene graphs in most research works are generated by modeling using context information among targets, focusing only on the inference process but ignoring the integrity of the input target information and the impact of the global information of the targets on relationship inference. Therefore, a new scene graph generation method based on global embedding and contextual fusion (GECF) is proposed in this paper. In this method, richer entity information is obtained by embedding global information into the entity features, while more robust inference of entity interaction information and more reasonable relationship fusion are acquired by combining the attention weighting module and the context inference module as a joint inference module, and merging the obtained entity features according to their discrepancy. The experiment on Visual Genome dataset shows that GECF method performs better than the existing methods in scene graph visual relationship detection.

To reduce noise and background distractions, we propose a detection algorithm based of Gaussian curvature filtering (GCF) and partial sum of singular values (PSSV). Above all, aiming at the false alarms caused by noise, using the prior knowledge of natural image have the characteristics of approximate developable, GCF utilizes the variation model to obtain an approximately noise-free image. Secondly, we adopt PSSV to suppress the background by pointing out that the reason why Robust PCA is inaccurate in the background estimation is that the hypothesis at the edge of the background does not match the reality, and we present a model solving algorithm on the ground of imprecise augmented Lagrangian multiplier method (IALM). At last, we use an adaptive threshold segmentation algorithm to segment the target. The model has better noise clutter suppression and detection accuracy than other representative methods.

Noticing that the human motion in both the positive and inverted time sequence are predictable, we design a reciprocal network which contains a forward and a backward prediction network to learn this information respectively, and then train the two networks in a collaborative and iterative way with the guidance of consistency constraint. Based on the reciprocal network, we establish an error compensation network to approximate the error of forward prediction network for making the forecast closer to the real pose. Extensive experiments show that our approach outperforms most recent methods in both short and long term motion predictions on Human 3.6M and CMU Mocap.

Body-worn cameras (BWCs) have been widely used all over the world in recent ten years. China is one of the earliest countries to deploy BWC system. In China, BWC has evolved from recording equipment for law enforcement to service BWC, working BWC and so on. Not only can public security men use BWC system to complement Skynet of the Ministry of Public Security, even to command and dispatch in public security affairs processing, but also other mandatory or service industries can get digital evidence by utilizing the system. From a technical point of view, we summarize BWCs in China. The emerging new wireless technology has driven BWC systems from single hardware equipment to systemization. Docking station is introduced and platform has evolved fast. To support more various services of transmission BWCs is the future direction of the development of BWC system.

Among structured light strategies, phase measuring profilometry (PMP) is considered to be an accurate and robust 3D reconstruction method. However, in practical scenes, gamma distortion is an inevitable factor that affects the accuracy and resolution of the final measurement result. In this paper, we propose a square-wave encoding scheme, which is gammafree, and we also demonstrate why the scheme is immune to gamma distortion. Then, based on the epipolar geometry, the invalid phase is removed by the epipolar constraint. Experiments demonstrate that the method can effectively reconstruct the surface of the scanned object and remove the invalid phase. With the proposed method, the accuracy for reconstructing a vertical plane is effectively improved. The method can be an effective alternative in the gamma distorted systems.

Images are one of the important carriers for acquiring underwater information. This paper proposes an underwater image enhancement algorithm with multi-input fusion to address the problems of low image contrast and colour degradation caused by the underwater environment in the imaging process of underwater optical images. Firstly, the dark channel a priori defogging is used to remove the inhomogeneous turbidity in the image and equalise the image color; then, the restricted contrast adaptive histogram equalisation is used to uniformly clip the image brightness components to further improve the image contrast; at the same time, the homomorphic filtering is combined to solve the problems such as uneven illumination of the image. Finally, the multiple outputs obtained are weighted using the Euclidean rule to obtain the final fused image. Experiments show that the images enhanced by this method are clearer to the naked eye and the target information is more prominent.

The application of abnormal event detection in video surveillance is an active research field, but due to the imbalance of positive and negative samples in surveillance video, abnormal event detection is full of challenges. In this paper, we propose a new abnormal event detection method based on appearance repair and motion consistency for detecting anomalous events. Specifically, the input image is partially masked and then fed into our proposed appearance repair autoencoder for image reconstruction, and then the motion consistency of images is constructed by our proposed optical flow network. The experimental results on the UCSD, CUHK Avenue datasets show the superiority of the detection performance of our method.

In this paper, we propose an end-to-end neural network abbreviated as TCNN to solve the blind phase retrieval problem in multiple scattering imaging. TCNN is a kind of auto-encoder with a transform layer, which acts as a bridge between transforming domains. Compared to double phase retrieval method, TCNN can directly estimate the image from those phaseless measurements through the nonlinear network structure. During training, the parameters of TCNN are updated by the adaptive moment estimation algorithm Adam. Numerical experiments show that TCNN can recover images with comparable quality to that of state-of-the-art methods. Moreover, TCNN hugely reduces the time cost for recovering images once the training procedure is completed.

In recent years, binary descriptors have attracted more and more attention due to their low memory consumption and high speed. It is well known that these representations are worse than higher-dimensional and histogram-based descriptors such as SIFT. Therefore, this paper proposes a fusion gradient distinction binary image descriptor (GDBID). Gradient comparison is added on the basis of the original gray comparison to enrich the information contained in the descriptor. At the same time, the comparison patches of different sizes are obtained by constructing concentric circles to achieve anti-noise. In addition, a threshold is set to filter patches to reduce the dimension of descriptors. Experimental results show that the GDBID has a precision is close to the best algorithm (SIFT), and the time consumption is lower than the fastest ORB in the literature.

The inefficiency of the dimension measurement on express packaging box lags behind the fast development of logistics industry in China. Therefore, focusing on improving the measurement efficiency of express packaging box, this paper proposes a new study on combining dimension binocular camera together with artificial intelligence algorithms. Specifically, a method combining HSV color space with OTSU is proposed to extract the box contour, straight line fitting is used to extract box corner points, and epipolar constraint is used for stereo matching of images, and finally the box dimension is obtained. The results show that in the experimental condition, this method can effectively control the relative error rate of the measurement of express box dimension to about ± 5 mm, which is close to the theoretical precision.

With the rapid development of economy, the number of vehicles has increased sharply, and the parking problem has become increasingly prominent. By building a parking sharing system, resources are integrated for rational use. Cloud technology, NB-iot communication technology and image recognition technology are used to realize the interconnection of vehicles, parking equipment and parking lots, and increase the functions of reservation and access vehicles, AGV parking guidance, mobile payment, parking information query and so on, so as to seek reasonable and efficient utilization of parking resources and alleviate the problem of difficult parking.

The outbreak of Covid-19 in specific areas characterized with global industrial chain reconstruction, grasps the epidemic prevention and control, while the growth of economics is the key point of government management in the post-epidemic time scale. In the last two years, various industries were affected by the outbreak of the new champions league experience “determination to keep factories and workers idle - to return to work and production - not to return to work and production”, which significantly changes the industrial performances. Both industry change and the development of the pattern of economic evaluation are equally important. This study used media platform composite index and edge parallel coordinates visualization method of combining the bundle, selected in 2020 and 2021, 11 industry information volume in 7 platform, through the medium of agenda building industry change from comparative analysis of two developing and development situation. This paper indicated that the major factors influencing the change of media attention in the course of the epidemic were the degree of social risk and the basic needs of the public. The epidemic had a great impact on the development of all industries, and the continuous improvement of the industry’s ability to deal with major risks became the main trend of future development.