This PDF file contains the front matter associated with SPIE Proceedings Volume 13509, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

Light fields can simultaneously capture the intensity and direction of each light ray, offering enhanced information for depth estimation. Currently, accurately extracting the epipolar plane lines and handling occlusion are the tough problems of the epipolar plane image depth estimation for light field. Passive depth estimation algorithms only relying on texture information from reconstructed object surfaces struggle to reconstruct texture-free areas, and exhibit poor noise resilience and low accuracy. To address these problems, this paper introduces structured light projection to add accurate and stable texture information, thereby improving the measurement accuracy and robustness, and proposes an epipolar plane line extraction algorithm suitable for structured light field data. This paper designs a spinning line operator based on Structured Light Field characteristics, and uses the variance of the points on the epipolar plane line as the cost of measuring the extraction accuracy. It judges and handles the occlusion problem by gradually removing abnormal points to make the variance converge. This paper measures and reconstructs objects such as standard blocks, standard balls, and portraits, and achieves satisfactory reconstruction results.

Image steganography plays a significant role in safeguarding privacy, security and confidential documents. Although current deep learning algorithms associated with this technology have reached a high level of safety, their application in 3D information encryption remains limited. As diffusion models substantially impact image generation, in this paper, we propose an approach to encrypt optical fringe images using the DiffStega method, one of the diffusion models. Because a fringe image carries the height information, the 3D information can be hidden and transferred through a confidential image. Experiments show the effect of our method, which can encrypt and recover the fringe image successfully.

Fringe projection profilometry (FPP) is one of the most popular 3D sensing techniques and is widely used due to its flexibility, high resolution and high speed. However, the conventional FPP has errors in absolute phase and 3D reconstruction results when measuring the surfaces of high-reflective objects because the fringe images contain some overexposed areas. To solve this problem, a two-wavelength polarization-encoded structured light method is proposed in this paper, which takes advantages of the properties of commercial liquid crystal display (LCD) projector to modulate the intensity and polarization state of the projected light simultaneously, and use two mutually perpendicular polarization states to project two fringe patterns with different frequencies. The polarization camera can capture them in separate polarization channels to achieve the desired phase unwrapping. The proposed method enhances the stability of the fringes, reduces the required fringe pattern by half, and increases the measurement speed. Experimental results show that the proposed method can suppress the phase error in the highlighted region when reconstructing high-reflective surfaces, obtain accurate absolute phase and reliably perform 3D reconstruction.

This paper introduces a visual odometry scheme based on a semi-feature method aimed at enhancing the efficiency of front-end tracking in visual SLAM systems. By integrating direct and feature-based methods, our approach improves tracking speed while maintaining pose estimation accuracy. We replace traditional pose estimation with a direct method for regular frames and utilize feature-based methods for keyframes, optimizing computational resources without significant loss in performance. Experimental validation on the EuRoC dataset shows substantial improvements in real-time tracking efficiency compared to conventional feature-based methods.

Feature extraction is a critical step in point cloud registration networks, determining how effectively point cloud data is represented. The feature extraction module in RPMNet uses fixed-range local features as input, whereas DGCNN dynamically constructs adjacency graphs and uses graph convolutions for feature updates, better capturing both local and global features of the point cloud. This study aims to explore the application of the Dynamic Graph Convolutional Neural Network (DGCNN) in RPMNet’s feature module to address the inadequacies in local and global feature extraction and improve its registration accuracy in point cloud data processing, particularly on the ModelNet40 dataset. The study conducted training and testing on the ModelNet40 dataset, comprising 5124 training point clouds, 1198 validation point clouds, and 1245 test point clouds. Through performance comparison analysis, various evaluation metrics such as rotation error and translation error were used to assess the model’s performance. The results show that the DGCNN-enhanced RPMNet reduced the isotropic rotation error from 0.056 to 0.053. This indicates that applying DGCNN to RPMNet can dynamically capture local and global features of point clouds, improving feature representation accuracy and model robustness. These findings are significant for the field of point cloud data processing, validating the effectiveness of DGCNN and providing new directions for future research. This advancement promotes the application of graph neural networks in practical problems, enhancing the technical level of the related field.

With the rapid development of 3D reconstruction, point cloud registration technology, a key step in 3D data processing, has garnered significant attention. Existing 3D point cloud registration technologies face issues such as low matching rates in coarse registration, misidentification of feature points, lengthy registration times, and low registration accuracy. Consequently, an improved registration algorithm that combines Fast Point Feature Histogram (FPFH) for coarse registration and Colored Iterative Closest Point (ColorICP) for fine registration has been proposed. The process begins with pre-processing the point cloud through downsampling filtration; next, point cloud features are extracted using FPFH to achieve feature matching and obtain the initial transformation matrix; finally, ColorICP is used for fine point cloud registration. Experiments on the Stanford bunny dataset from the standard point cloud library and real-world point cloud data demonstrate that the proposed registration algorithm effectively utilizes both color and geometric features of the point cloud, achieving significant improvements in registration accuracy and duration compared to traditional registration algorithms.

The precise and efficient reconstruction of three-dimensional structures stands as a pivotal cornerstone for the automated management of fruit trees. This study introduces a three-dimensional reconstruction methodology for fruit trees, leveraging incremental analytical scene motion recovery technology, alongside a tailored data acquisition framework for characteristic fruit trees in Xinjiang. By harnessing visual sensors, envelope image data of fruit trees is captured to form a panoramic multi-view image dataset. Key frames are extracted to derive a multi-view representation of the fruit tree, and incremental analytical scene motion recovery technology is employed to train pose information and generate a point cloud model, yielding a high-precision three-dimensional representation of the fruit tree. During experimentation, error analysis focused on the five key diameters of the target fruit trees, revealing reconstruction deviations ranging from 2.43mm to 5.24mm, with reconstruction errors falling between 4.31% and 5.93%. These findings underscore the methodology's capacity to strike a delicate balance between reconstruction precision and cost-effectiveness, offering enhanced technical underpinnings for the automated management of orchards.

Traditional multi-view three-dimensional (3D) reconstruction uses images from visible light that may be of poor quality under conditions such as reflections, light spots, or haze, causing difficulties for 3D reconstruction in large scenes. To solve this problem, an efficient hybrid structure in motion (SFM) and multi-view stereo (MVS) point cloud reconstruction and alignment fusion method based on infrared images is proposed in this paper. Since thermal infrared imaging is less affected by lighting conditions and smoke, we adopt infrared images as the reconstruction raw dataset to make up for the defects of visible light 3D reconstruction; the reconstruction method adopts the hybrid SFM-MVS algorithm reconstruction with subset partitioning strategy, which can complete the dense reconstruction more efficiently under a large amount of image data. The iterative closest point (ICP) algorithm is used to perform the fusion of multi-angle reconstruction for the alignment of the point cloud, which can make up for the lack of side information in the large scene model reconstructed from a single angle. The experiment finally obtained a fairly good infrared 3D scene model with complete side information.

Traditional fringe projection profilometry (FPP) struggles with robust imaging and high-precision three-dimensional (3D) reconstruction in complex lighting environments with strong inter-reflection. This paper proposes a method combining frequency-shifted fringe projection and epipolar constraint to achieve high-precision 3D profilometry in such scenarios. First, environmental light interference is mitigated and the maximum fringe frequency is reduced through window function analysis (WFA) and a four-step phase-shifting (PS). Then, direct and indirect illumination are separated in the Fourier domain. Direct illumination positions are identified using threshold segmentation and epipolar geometry constraint, enhancing imaging robustness in strong inter-reflection contexts. Experimental results demonstrate that the method constructs direct and indirect illumination feature models, accurately obtaining direct illumination information and eliminating multi-path interference. Additionally, in tests with complex geometrical surfaces, imaging efficiency and robustness significantly improve, with effective point cloud data increasing by over 30% and planar accuracy improving by 50%.

Accurate and comprehensive three-dimensional (3D) measurement of intricate microstructures remains a significant challenge. This paper presents a 3D imaging method based on multi-view projection. To address the challenges of limited depth of field (DOF) and occlusion in small field of view (FOV) measurements. We use a multi-view setup consisting of one telecentric camera and four Scheimpflug projectors. To simplify the complex computational process involved in system calibration, we developed a model based on the phase-to-height mapping relationship under complex optical paths, which also enhances the calibration accuracy of the system. The 3D reconstruction results demonstrate that this method achieves good performance and precision for small FOV 3D imaging, meeting the requirements for measuring microstructures.

As the characteristic parameters of multi-sensing keyhole and plume reflect effectively the welding quality of high power disc lasers, high power disc laser welding of high strength steel SX780CF was used as the test object, and the dynamic image of molten pool during welding was extracted simultaneously by high speed camera and photoelectric sensor vision imaging system. The images of molten pool and metal vapor obtained by high-speed camera in laser welding experiment were processed by OpenCV. The characteristic parameters of keyhole and metal vapor are extracted by means of median filtering, binary thresholding and open operation. The relationship among keyhole shape, welding condition and welding state was explored by analyzing the variation law of keyhole area and metal vapor area curve. Two characteristic parameters of visible light intensity and reflected light intensity during laser welding were obtained by photoelectric sensor. The change law of visible light and reflected light curve is analyzed, the power spectral density of photoelectric signal is explored, and then the welding state in the welding process is judged. The on-line monitoring of welding process is realized.

Coherent beam combining (CBC) technology increases the peak power of laser systems by combining multiple coherent beams, overcoming the power limitations of single fiber lasers. This technology is essential for high-power laser applications in military and medical fields. Precise phase control is crucial for efficient beam combining, and reinforcement learning (RL) methods are being explored to address this challenge. This paper introduces a deep RL inference framework that supports various agents and includes an adaptive retraining mechanism to maintain performance amid changing environmental conditions. When a 90% drop in performance is detected, the framework triggers retraining using the improved proximal policy optimization (PPO) algorithm until performance recovers to 95%. Simulations show the framework can adapt to new environments within 1-3 training cycles, ensuring continuous learning and stable performance in dynamic settings.

Structural settlement is one of the most critical indicators in structural health monitoring. In long-term vision-based measurement, the stability of the observation platform is a major bottleneck limiting its application. Due to the optical lever effect, a slight angular change in the camera during long-distance measurements can result in significant errors, even leading to measurement failures. This paper proposes a vision-based self-calibrating measurement method using gravity balance for field applications. In this method, the main camera, sub-camera, and a plumb line are rigidly connected. The sub-camera tracks the plumb line to obtain the angular changes of the main camera. The rigid connection between the main and sub-cameras is used for angle compensation, enabling self-calibrating measurements on unstable observation platforms. Experimental results demonstrate the accuracy and effectiveness of the proposed method. This method has the potential to extend the application of visual measurement in monitoring large-scale engineering structural deformations.

A surface plasmon sensor (SPR) utilizing Ti₃C₂T_x nanosheets is proposed in this work. The findings reveal that sensitivity increases initially and then decreases with the number of deposition cycles. A maximum sensitivity of 3467.8nm RIU⁻¹ is achieved after one deposition cycle, marking a 71.27% improvement over the unmodified sensors. In addition, the shifts in resonant wavelength in-creased from 2.89nm to 6.35nm in the test with a 0.4mg/mL Bovine Serum Albumin (BSA) solution. These results suggest that the Ti₃C₂T_x modification can significantly enhance the sensitivity of SPR biosensor.

The scattering media measurement for structured light techniques remains challenging due to the subsurface scattering that destroys fringe patterns and introduces phase error. Recently, deep learning has been successfully applied to fringe analysis. Unfortunately, collecting a large scale of labeled data is difficult, especially in scattering media measurement scenarios. Inspired by image style transfer, we proposed a high-precision scattering medium measurement method based on self-supervised fringe domain transformation. Specifically, by treating degraded fringe and ideal fringe as two pattern styles, a cycle generation strategy is developed to achieve fringe enhancement by translating between two fringe domains. Two generators and two discriminators are used to form cycle-consistency constraints for self-supervised learning with unpaired dataset. Additionally, we combine numerical and physical constraints to jointly optimize network parameters, effectively suppressing phase-shift errors and improving phase quality. Since the proposed approach eliminates the need for label data, it exhibits great potential for applying deep learning to scattering medium measurement. Experiments verify that our method can obtain high-quality 3D morphology of translucent objects.

Aiming at the problem that the monocular digital image correlation (DIC) system can only measure the information within the object surface, a monocular DIC system is proposed to realize 3D panoramic measurement. That is, using the principle of multi-plane mirror imaging, the front and rear surfaces of the measured object are imaged in a common field of view, and the two optical paths are imaged in two plane mirrors respectively, which extends the monocular into four virtual cameras to realize 3D panoramic measurement. 3D reconstruction and comparative experiments verify the feasibility and accuracy of the method. The results show that the 3D reconstruction of the front and back surfaces of the coin is performed, and the reconstruction results are consistent with the morphology of the decision coin.

The structure and material of a musical instrument decisively influence its vibration modes. In turn, the precise measurement of vibration aiding in analyzing the structural and material properties of the musical instrument, and ultimately provides a basis for evaluating the performance of the musical instrument. This paper leverages the superiority of the digital speckle pattern shearing interferometry which also named digital shearography in optical dynamic measurements to determine the vibration distribution of musical instruments. The combination of digital shearography and time-average method is used to qualitatively determine the surface vibration modals of a musical instrument under different orders of intrinsic frequency excitation. The phase shift technique introduced by the time-average method removes the straight flow to improve contrast and make the modal streaks on the surface of the instrument clearer. Different modes of image subtraction are modulated by various expressions of the Bessel function, reflected in the changes of brightness and darkness of the fringes. Furthermore, in light of the properties of shearography within the field of non-destructive testing, this technique is adept at detecting flaws in the structural integrity of the instrument itself. Such deficiencies may encompass variations in the uniformity of surface material composition, the existence of cracks, irregularities in solder joints, and other associated concerns. This research expands the application of optical dynamic measurements in analyzing musical instrument vibrations and provides theoretical and methodological support for future research, thus introducing new technological tools to the industry.

Urban road settlement is one of the key indicators for its safety warning. In the project, the road settlement observation point consists of a hollow cylindrical protective sleeve and a settlement rod on the center axis of the cylinder. This paper proposes an urban road settlement measurement method based on optical images. Firstly, the ellipse detection algorithm is used to identify the circular orifice features of the settlement monitoring point, and the angle between the camera and the road surface is recovered by the pixel size information of the ellipse long and short axes; secondly, the template matching algorithm is used to sub-pixel locate the diagonal features at the top of the settlement bar; finally, the settlement measurement model is established. On this basis, an urban road settlement measurement application (Easy-to-Go) based on Android platform was developed. The experimental results show that the measurement accuracy of the proposed method is better than 1mm, which meets the engineering measurement requirements. The research results of this paper can provide a convenient, low-cost and reasonable accuracy new method for road settlement measurement.

Laser Doppler Vibrometers (LDV) have garnered widespread attention and application across various fields due to their high-precision non-contact measurement capabilities. This study employs a single-frequency laser at a near-infrared wavelength of 1064nm as the light source and constructs an external heterodyne laser interferometer system based on a Mach-Zehnder interferometer. By modulating the reference light with a 73 MHz acousto-optic modulator and causing interference with the measurement light on a photodetector, the vibration signal is extracted using phase-locked loop demodulation technology. In the experimental part, we verified that the constructed LDV system can effectively reproduce the 20Hz to 20kHz sinusoidal electrical signals applied to the vibration target, proving the system's wideband response capability and high sensitivity. In addition, by collecting audio signals played by a speaker and applying noise processing algorithms, clear voice signals were successfully restored from the collected data. This indicates that LDV technology can not only monitor vibrations from low to high frequencies but also effectively recognize voice signals in practical environments. This study explores the application potential of LDV in the field of remote voice monitoring. The research results provide a new perspective for remote voice monitoring and recognition and offer valuable references for the development and practical application of related technologies and practices.

With the development of urban underground spaces, ensuring the safety of operational subway tunnels poses significant challenges. Among these challenges, tunnel structural deformation and foreign object detection are crucial components of safety protection. Visual measurement offers specific advantages such as non-contact operation, all-weather functionality, high precision, and real-time dynamics. This paper proposes a method based on computer vision for detecting structural deformation and foreign objects in operational subway tunnels. Firstly, current tunnel morphology data is obtained through monitoring cameras. Then, digital image processing techniques such as image denoising, feature segmentation, sub-pixel positioning, filtering, and displacement calculation are applied. By comparing the processed sequential data, deformation data such as subsidence changes and horizontal changes are obtained, and the presence of foreign objects is determined. Experimental results indicate that the proposed method can simultaneously identify tunnel structural deformation and foreign object intrusion. This addresses the limitations of existing technologies, which have singular monitoring functions and incomplete tunnel health assessment systems, thereby ensuring the safety of subway operations.

The accuracy of image feature location is one of the key factors affecting the accuracy of optical measurement. Circular diagonal markers are commonly used cooperative features in optical measurement. In complex field imaging environments, factors such as spatial illumination, environmental noise, and camera shaking will cause problems such as uneven brightness, incomplete, overlap, and image blur, which will affect the locating accuracy of markers. To solve the above problems, this paper proposes a sub-pixel detection and localization method for circular diagonal markers in complex field environments. Firstly, the double-platform histogram equalization is used to enhance the image to solve the uneven brightness phenomenon. Then, Hough circle detection is used to define regions of interest around circular markers. Subsequently, a combination of multi-scale and multi-target template matching and surface fitting extremum methods is utilized to achieve high-precision sub-pixel detection and localization of markers, even when they are incomplete, occluded, or blurred. The experimental results show that the proposed method can accurately locate all circular diagonal markers in complex field environment. Compared with the existing methods, the proposed method has more accurate localization precision and has the advantages of automation, strong robustness and wide applicability.

The slip accumulation induces microscale plastic localization, often leading to irreversible damage in nickel-based single-crystal superalloys. While most studies have quantitatively described plastic localization through numerical simulations, few experiments have characterized the microscale inhomogeneous strain distribution of slip damage. In this study, we employed the sampling moiré method to measure the sub-microscale strain and displacement fields in nickelbased single-crystal superalloys under tensile loading. Our analysis quantitatively assessed the formation of slip traces and the evolution of slip bands during mechanical loading, providing valuable insights into the mechanisms of slip damage. The sampling moiré method can obtain sufficient high spatial resolution by reducing the pitch of the deformation carrier grids, demonstrating its superiority in microscale deformation measurements and its potential for characterizing the mechanical behavior of materials.

This paper introduces a novel 3D method using a multi-channel Laser Doppler vibrometer (LDV) to measure axial and radial error motions in high-speed rotating motor shafts. The approach utilizes three orthogonal LDVs to accurately capture 3D displacements from the shaft's surface. This enables the measurement of error motions along both the axial (sensitive to motion parallel to the shaft) and radial (sensitive to motion perpendicular to the shaft) directions. The collected rotational motion data undergoes post-processing to obtain the synchronous and asynchronous components of the axial and radial error motions. Experimental results demonstrate that the shaft's rotational speed significantly influences error motions in both the axial and radial directions, as evidenced by measurements at different speeds. Moreover, analyzing the data facilitates detecting potential damage in the rotating shaft. Finally, an analysis quantifies the combined overall uncertainty in error measurements using this new method.

Structural vertical displacement, lateral displacement, and torsional deformation are important indicators for the health monitoring of long linear engineering structures such as bridges and tunnels, but there is currently a lack of practical engineering monitoring methods and technologies for multi-parameter collaborative measurement. In this paper, an in-plane rotation-angle-relay videometric model for torsional deformation of long structures is first established based on the perspective projection. Furthermore, a new method of serial camera network videometric for collaborative relay of in-plane rotation angle and displacement of long structures is proposed by using an optical target that fuses point and line feature information and combining it with the classic displacement-relay videometric method. The effectiveness and engineering practicality of the proposed method were verified through simulation experiments and laboratory-scale physical experiments. The experimental results show that the method proposed in this paper can achieve the angular minutes-grade (1.44 angular minutes) measurement of the torsional deformation of long structures, even when the measurement platform itself is unstable. In addition, the measurement results of vertical and lateral displacements can be corrected for the influence of platform shaking, with a correction effect of up to 86.75%. This validates the high-precision measurement of displacement and rotation angle that can be achieved with the proposed method even when there is shaking of the measurement platform, demonstrating strong potential for engineering practicality.

In this paper, we present a consistency analysis for focal stack-based light field reconstruction. A pre-calibrated light field camera is used to construct the metric mapping relationship in image-object space and the reference light field, based on which light fields are reconstructed and quantitatively analyzed. Sequentially, a consistent light field reconstruction method is proposed. Image-object space transformation is established through camera calibration and image distortion is corrected simultaneously in the focal scanning space. The image stack can then be uniformly converted from image space to object space for consistent light field reconstruction, which can finally realize depth measurement in the focal scanning space.

Smart agriculture leverages information technology and IoT to achieve automation, intelligence, and precision in agricultural production. Harvesting robots, as crucial components, must accurately understand orchard environments and crop damage to optimize crop management and harvesting strategies, enabling efficient, low-loss, and large-scale operations. However, current semantic segmentation models often fail to meet the high-efficiency requirements in orchards, particularly in segmenting small-pixel targets, resulting in misidentification, missed detection, and segmentation distortion. To address these issues, we propose an improved segmentation network, BoT-PSPNet, for precise segmentation of orchard roads, trunks, fruits, and their damage. Specifically, we use PSPNet as the primary framework and integrate the Multi-Head Self-Attention (MHSA) module from Transformer into the ResNet50 Bottleneck. This approach overcomes PSPNet's limitations in handling long-range dependencies. The MHSA module provides robust global feature extraction capabilities and, through its self-attention mechanism, effectively captures long-range dependencies, addressing PSPNet's deficiencies in multi-scale information fusion, computational efficiency, and flexibility. We collected a dataset of 1,000 orchard images, including road conditions and various fruit damages, and conducted validation experiments. The results demonstrate that the proposed method outperforms traditional PSPNet models in segmentation accuracy, particularly in handling complex backgrounds and detailed features. This study provides a new methodological support for enhancing the visual perception accuracy of scenes and targets in agricultural environments.

To solve the problem of low recognition accuracy of outdoor orchard, it is greatly affected by outdoor light and other environments. In this paper, based on the deep learning neural network YOLOv9 algorithm, the cherry tomatoes in the outdoor orchard are accurately identified. In this paper, YOLOV5, YOLOV7, YOLOv8 YOLOV9, and YOLOv10 algorithms were used to detect mature cherry tomato fruits, immature cherry tomato fruits and damaged cherry tomato fruits. Compared with other YOLO series algorithms, YOLOv10 algorithm has higher detection accuracy and is less affected by outdoor light. Through the comparative test analysis, the mAP value of the model is 82.34%, the F1 value is 79.91%, the accuracy is 86.82%, the recall rate is 74.02%, the detection rate is 30.3 frames/s, and the model volume is 15.97 MB. It meets the requirements of outdoor detection and can provide identification basis for outdoor cherry tomato picking technology.

In complex tea plantation environments, efficient navigation path recognition algorithms are essential to guarantee the autonomous operation of mobile harvesting robots. However, the relevant path recognition algorithms are poor in accuracy and real-time, making it difficult to identify complex tea plantation paths quickly and accurately, which affects the efficiency and robustness for autonomous operation of tea plantation mobile harvesting robots. To address the above problems, firstly, this paper selected Deeplabv3plus as the model framework for path recognition, and employed lightweight Mobilenetv2 as the backbone network of the model to optimize the number of arithmetic parameters and the speed of the model. Secondly, an efficient CE_ASPP feature extraction module was designed to construct a multi-scale depth-separable convolutional structure to expand the feature receptive field of the model. In the decoding layer, this paper designs the CMG feature extraction module, which integrates the shallow features of the first three bottleneck layers of the backbone network, is fused to enhance the semantic information of the features; and the feature extraction capability was improved by the CBAM attention mechanism and the ECA attention mechanism. Finally, the scanning method was employed to extract the central pixel, while the RANSAC algorithm facilitates the fitting of the navigation path. The experimental results showed that the Accuracy, mPA and MIoU of the improved Deeplabv3plus model were 95.7 %, 97.64 % and 93.51 %, respectively. Compared to the original model, the number of operational parameters and memory footprint of the improved model were 4.35×10⁵ and 16.60MB respectively, which significantly reduces the computational costs. The average inference time and the average detection frame rate were 48.99ms and 20.44FPS respectively. The results showed that the improved Deeplabv3plus has better detection accuracy and robustness for complex tea plantation paths, providing important solutions for autonomous operations of tea plantation mobile harvesting robots.

In robotic harvesting, recognizing and locating target apple fruits, as well as detecting obstacles in complex environments like dense canopies, poses significant challenges. This paper presents a method for recognizing, segmenting, and accurately localizing obscured fruits. We employ instance segmentation to extract edge contours of fruits and branches, leveraging two-dimensional features for precise localization under occlusion and correcting picking point errors. Experimental results indicate that at a 10% occlusion ratio, the model achieves an average precision of 96.8%, an F1 Score for edge detection of 0.91, and a 3D reconstruction error of 3.1mm. Even at 50% occlusion, average precision remains at 88.3%, with an F1 Score of 0.80 and a reconstruction error of 7.6mm. These findings demonstrate that while occlusion significantly impacts 3D reconstruction, the algorithm retains high accuracy, fulfilling the precision requirements for fruit harvesting.

Aiming at the problems of low fidelity, long time-consuming and high cost of constructing fruit tree models in the virtual orchard scene, this paper proposes a 3D reconstruction method of virtual fruit trees based on Structure from motion-Multi-view stereo (SFM-MVS). First, the fruit tree image acquisition is carried out by using camera, the SFM algorithm is used to calculate the camera parameters of the fruit tree pictures and the positional relationship between the cameras, the image segmentation is carried out by combining the Convolutional Neural Networks (CNN) of the deep learning, and the segmentation of the fruit tree and the background of the environment in the image is completed by using the DeepLab algorithm. Secondly, the MVS algorithm is used to fuse the segmented fruit tree information and the associated camera position information to automatically construct a high-precision 3D model of the fruit tree. Finally, the mesh information and texture mapping of the 3D model are imported into the Unity3D virtual simulation platform, and the attribute fusion is realized by Albedo, which realizes the rapid digital model construction of real fruit trees.

In order to enhance the precision and efficiency of the assembly process for intelligent agricultural equipment, this paper integrates machine vision perception and deep learning algorithms, and proposes an OpenPose-based virtual assembly method for key components of equipment. This marks the inaugural instance of employing a binocular stereo vision system in conjunction with the Multiview Bootstrapping algorithm in OpenPose to address the challenges of hand positioning and occlusion during the assembly process. This approach enables the precise detection of hand key points and the accurate calculation of hand position information. To address the high cost of production of physical agricultural equipment, a digital twin model of agricultural equipment components and a virtual hand model are constructed. A virtual assembly platform is then constructed and applied to verify the visual system and conduct interactive virtual assembly tests. The results demonstrate that the platform can rapidly and accurately track the key points of the hand, thereby enabling the interactive virtual assembly of key components of agricultural machinery.

During the operation of long-staple cotton picking robots, the slight wobbling of cotton bolls caused by dynamic uncertainties can result in the loss of shape and texture information within a few pixels. Low signal-to-noise ratio, background clutter, and occlusion issues further degrade the performance of traditional detection methods, increasing the probability of missed detections and inaccurate positioning. This study proposes a neural visual pathway model based on the drosophila visual system for precise localization and dynamic tracking of cotton bolls. The model simulates the sensitivity of the drosophila neural visual pathway by responding to weak motions of small targets in cluttered backgrounds, and introduces a direction-selective inhibition algorithm to reduce background interference. Experiments show that the model performs stably in detecting small targets with a wide range of speeds and sizes, and can quickly and accurately extract motion direction and energy information.

Nighttime detection and harvesting are key issues for achieving all-day operation of tomato-picking robots. Currently, most general detection algorithms are limited to natural daylight conditions, with significantly reduced performance in nighttime environments. To address the issues of low accuracy and poor robustness of nighttime tomato detection algorithms, a high-precision nighttime tomato detection method based on the integration of deep learning and image processing is proposed. This study designed multiple sets of nighttime RGB lighting experiments to calculate the HSV color distance between ripe tomatoes and the background under each lighting condition, determining the optimal lighting color to enhance the contrast between tomatoes and the background. Under the optimal lighting conditions, an RGB image dataset of nighttime tomatoes was constructed, and an YOLOv8-based nighttime detection model was trained to achieve precise detection and localization of nighttime tomato targets. Within the detected target frames of ripe tomatoes, image processing methods such as OTSU, Hough detection, and connected component analysis were used to judge and analyze the occlusion situation of tomatoes, distinguishing between occlusion types (leaves or branches), and providing guidance for optimizing the robot's picking strategy. Finally, this study verifies the effectiveness of the algorithm through multiple sets of experiments. The algorithm has an overall accuracy rate of 84% and can be deployed on edge devices to achieve efficient real-time detection tasks while ensuring performance.

To address the challenges of inadequate visual recognition accuracy, sluggish path planning, and imprecise trajectory tracking that contribute to picking failures in multi-degree-of-freedom robotic arms, this paper introduces a visually guided trajectory planning and control strategy specifically designed for six-axis robotic manipulators. The proposed approach emphasizes effective obstacle avoidance and precise target tracking within complex and dynamic environments. The strategy is initiated by integrating the YOLOv8 deep learning-based visual detection model, which ensures rapid and accurate target localization on the pixel plane, thereby enabling crucial real-time object recognition. The visually detected targets are subsequently utilized to steer an enhanced RRT*-connect algorithm, wherein path search efficiency is significantly improved by implementing adaptive step size and intelligent sampling strategies, thereby minimizing the computational burden and the number of sampling nodes. To further optimize the planned trajectories, B-spline interpolation is employed to achieve smooth, continuous, and stable motion paths. In the final stage, a sliding mode controller integrated with Model Predictive Control (MPC) principles is developed to ensure precise and robust trajectory tracking. This controller exhibits a high degree of responsiveness, effectively adapting to dynamic changes and maintaining accurate path following despite the presence of external disturbances. Comprehensive simulations and experimental validations underscore the robustness and efficiency of the proposed methodology, highlighting substantial improvements in reducing the number of sampling nodes, accelerating convergence rates, enhancing path smoothness, and achieving superior tracking accuracy, making it well-suited for high-performance trajectory tracking tasks in complex and dynamic scenarios.

To reduce potential collisions and damage that may occur when harvesting multiple adjacent targets with a robotic harvester, this study proposes a grape harvesting sequence planning algorithm based on the removal of interference grape clusters and sampling distance adaptive clustering. First, the locations of grape clusters in distant scenes were identified using YOLOv8n. Subsequently, a method for removing background grape clusters and a distance adaptive clustering method for grapes were introduced, enabling adaptive harvesting cluster partitioning at varying sampling distances. Finally, a harvesting order planning algorithm for inter-cluster picking was proposed by integrating expert experience with the operational direction of the harvesting robot. Experimental results indicate that the distance adaptive clustering method demonstrated stronger adaptability for various sampling distances than the original DBSCAN clustering algorithm, and it is capable of grouping nearby grapes into a single picking cluster. The intra-cluster harvesting sequence planning algorithm effectively facilitated decision-making regarding the priority of grape picking cluster harvesting, aligning the harvesting sequence planning results with manually planned outcomes and providing essential information to guide the robot for low-damage harvesting.

With the increasing demand for high-speed mechanized operations in apple harvesting, the development of multi-arm harvesting robots for modern orchards has become a major trend in the harvesting robot industry. However, the level of agronomic standardization continues to present challenges to the advancement of existing multi-arm harvesting robots. Therefore, by analyzing the advantages and disadvantages of existing multi-arm harvesting robots and targeting the widely used Tall-Spindle apple orchards in modern China, this paper proposes a design scheme for a mobile four-arm apple harvesting robot. This design not only satisfies the demand for simultaneous four-arm picking in these scenarios, but also optimizes the allocation of tasks among the four cameras and robotic arms, thereby improving the overall picking efficiency. Experimental results indicate that the robot is capable of simultaneous four-arm harvesting in Tall-Spindle apple orchards. It successfully harvests 78.64% of the fruits within the growing range, and the average harvesting efficiency of the four arms is 3.39 fruits/s. This significantly improves the current challenges associated with the high labor intensity of manual picking.

In the process of automatic fruit harvesting, fruits can be obstructed by leaves and tree trunks, posing significant challenges to accurate and safe picking operation. This is due to the random distributions of fruits and positions of robots, and direct grasping of fruits by the robotic arm may lead to collisions with tree trunks and branches, resulting in malfunctions. To address the issue of collisions during the fruit picking and storage process, this paper proposes a target relative position estimation method based on deep learning-driven visual perception and a two-steps picking strategy to enhance safety during picking. This method initially employs a deep learning network to identify the locations of both fruits and tree trunks. Then, it analyzes the positional information of fruits relative to tree trunks to determine different picking and storage strategies, which includes the one-step method and two-steps method. Additionally, to minimise the time loss incurred during the two-steps picking process, this paper provided an optimised trajectory planning method for the robotic arm. The time costs of different path planning methods required to pick apples during planning and picking have been examined. Experimental results indicate that, for a target point, the RRTconnect algorithm is optimal in terms of time efficiency for path planning. Furthermore, the time required varies depending on the path taken by the robotic arm's end effector, while the shortest path may not necessarily correspond to the minimum time consumption.

The purpose of this study is to explore the application of virtual and real simulation technology in the control system of litchi picking robot. With the rapid development of agricultural automation and intelligence, picking robots are important equipment in modern agriculture, and the design and optimization of their control systems have become the focus of research. At present, the litchi picking robot control system has high technical requirements for operators, and the development cost is high and the risk is high. Secondly, the litchi picking robot is highly dependent on the real environment in the experiment and can easily cause damage to the natural environment. In view of the special environment and operational requirements of litchi picking, virtual simulation technology is used, combined with the performance characteristics of physical robots, to build a simulated litchi orchard scene and simulated picking manipulator in the virtual environment. At the same time, a real-time communication platform was built in Unity3D, and the digital twin control framework of the picking robot was creatively designed to achieve coordinated control and coordinated movement of the real and virtual robotic arms. The accuracy of the virtual-real combination control system was verified through design accuracy experiments. After experimental verification, the error rate of virtual and real motion of the robotic arm reaches less than 0.03%, which can meet the precise operation requirements of the robotic arm. The system can simulate the operation process of the picking robot in a virtual environment in advance. Experiments can also be performed without the need for actual equipment, reducing the risk and cost of development. In addition, the virtual orchard environment can also be used for smart agriculture research and lay the foundation for smart agriculture. It can also reduces dependence on the real environment and protects the ecological environment.

In this paper, a fringe projection profilometry (FPP) system for three-dimensional (3D) imaging of the human back is proposed to assess scoliosis, which has developed into one of the top four health concerns affecting adolescents. The FPP system includes a binocular stereo vision system and an infrared projector, with a Field Programmable Gate Array (FPGA) based synchronization control system enabling a 240Hz acquisition frame rate. This high frame rate mitigates the impact of involuntary body movement on 3D image quality. Various features can be extracted from the generated 3D model of the back for quantitative postural assessment. Specifically, a deep learning approach is employed to analyze the extracted trunk rotation angle curve for scoliosis detection. Experimental results demonstrate a correlation between this method and X-ray detection, offering a promising exploratory approach to reduce reliance on X-ray imaging in scoliosis monitoring.

This paper focuses on the field of educational games targeting children aged three to six, exploring the application of hologram in interactive card game design. It first examines the potential applications of holography in children's education, emphasizing the crucial role of interactive games in promoting children's cognitive development. Furthermore, the paper elaborates on the core principles of game design, including educational value, interactivity, entertainment, and safety, while highlighting the importance of considering children's psychological and cognitive characteristics in the design process. This study demonstrates the integration of hologram with children's educational games by introducing interactive mechanisms, visual design, narrative techniques, and educational objectives, creating game content that is both educational and entertaining. Through preliminary user testing, feedback was collected from child users and the game's impact on their cognitive abilities, creativity, and problem-solving skills was evaluated, with results supporting the potential of games in promoting children's development. Finally, the paper provides prospects for the future development of holographic interactive card games and proposes further design improvements and research directions based on testing feedback, aiming to bring innovative educational tools to the field of children's educational games.

This study focuses on human-pet emotional memories, discussing the potential application of experimental holographic art in this field. As holographic technology matures, it offers new opportunities for the representation of emotional memories in artistic creation. The research involves experimental art practices to analyze how holographic art impacts individuals’ memories and emotional experiences with pets. Initially, the study reviews the theoretical foundations of emotional memory and its expression in artistic creations. Furthermore, this paper introduces the creation process of interactive design of holographic art pieces. Additionally, an experiment is described where participants observe hologram of their pets. Their responses data are collected to examine the human-pet emotional memories. Finally, experimental results reveal the positive role of holograms in evoking emotional memories and enhancing emotional experiences, providing a prospect application in emotional education and memory healing for future.

This paper examines the system construction and the development of a novel talent cultivation model within the holographic arts center from an interdisciplinary perspective. It begins by discussing the importance of interdisciplinarity in contemporary education. The experimental and avant-garde nature of holographic arts center are also discussed in this context. It then explores the creation of a comprehensive and interactive experimental platform to facilitate deep collaboration and exchange among disciplines such as optics, design, and art. Additionally, it details the systematic construction for building holographic arts center, including basic technological infrastructure, innovative research platforms, creative spaces, academic exchange environments, industry linkages, and career trajectories for graduates. This paper proposes a novel Interdisciplinary I3 (Integration Interaction and Innovation) Model for talent cultivation. It is designed to enhance students’ abilities to integrate interdisciplinary knowledge, foster innovative thinking, and develop practical skills. It emphasizes the center’s role in providing practical opportunities, stimulating creative thinking, and advancing technological applications. The article also discusses historical models of integration between holographic artists and scientists, and the rise and fall of holographic arts courses, adapting historical insights to current educational methods to meet contemporary needs and characteristics. Furthermore, this paper analyzes challenges that may arise in constructing holographic arts center and cultivating talent, such as the technological change, the depth of discipline integration, and the teaching resource allocation, offering corresponding strategies for addressing these issues. Finally, it envisions the future development of holographic art laboratories as significant reference points for talent training and innovative research in the intersection of arts and technology.

This study investigates the emotional effects of varying display lighting conditions on viewer responses to holograms, focusing on how color temperature affects emotional reactions. An experimental design was implemented, where participants viewed the consistent hologram under diverse lighting conditions. Both qualitative and quantitative methods were conducted to generate participants’ emotional responses. The findings reveal that specific lighting conditions can influence these emotional responses. The analysis reveals that appropriate lighting enhances the holographic experience by modifying viewer emotional states. The results offer vital insights for optimizing holographic display design to improve viewer engagement and emotional experiences. Recommendations are presented for the strategic adjustment of display lighting to optimize the emotional impact, crucial for improving the efficacy of holographic displays within both entertainment and educational contexts. The study’s findings contribute significantly to the scholarly discourse by elucidating the crucial influence of display lighting on the emotional interactions experienced by viewers of holograms.

Installation art is a form that integrates various media and materials within a specific space. It transcends the constraints of traditional two-dimensional art form and evokes a multifaceted and emotional experience from viewers. Holography is a technique for recording and reproducing the wavefront of light, enabling the creation of three-dimensional images. Using holography in installation art broadens the perspective of artist creation. This paper explores the impact of natural light variations on the presentation of holographic art installations. The objective is to investigate how natural light can enhance the visual and emotional appeal of holographic installations while providing practical guidance for integrating this medium into artistic, educational, and environmental settings. By examining how physical holographic art installations manifest differently under dynamic natural light conditions and incorporating feedback from viewers, this study provides valuable insights for further exploring the integration of holographic art with natural environments. This research direction holds promise for offering artists, designers, and curators additional innovative design concepts and practical guidance.

War is a major issue of widespread societal concern, with its destructive impacts resonating deeply across the globe. Historically, numerous anti-war art pieces have not only delivered significant visual impact but have also provoked audience attention, reflection, and emotional resonance. Holography, as an innovative artistic medium, not only offers new ways of presenting content but also provides artists with a novel language for expression. This paper interprets historical holographic works on anti-war themes and explores the diversity and interactivity of this theme through experimental holographic artworks “Afterward” and “Aftermath”. These works provide viewers with new spatial and perceptual perspectives on the anti-war narrative.

As technology advances and cultural consumption evolves, museum cultural and creative products have increasingly become important vehicles for cultural dissemination and education. This study aims to explore the design of educational and engaging museum cultural and creative products using hologram. Initially, the paper outlines the principles and developmental history of holography in the field of designed product, analyzing its potential applications in museum cultural and creative products. Following market research and user needs analysis, the design direction and target user groups for cultural and creative products were identified. This study proposes a design scheme for holographic-based cultural and creative products, a Bronze Sitting Dragon from Capital Museum of China is used as a prototype for the design of holographic cultural and creative products, including aspects of functional modules, interaction methods, and visual presentation. A designed prototype is developed. The findings suggest that holograms can provide a new interactive experience and visual enjoyment for museum cultural and creative products, enhancing public awareness and interest in cultural heritage with broad application prospects.

Science-Technology Arts (Sci-Tech Arts) is an interdisciplinary field that plays a crucial role in stimulating human creativity and advancing technological innovation. As a burgeoning discipline within art education, Sci-Tech Arts profoundly influences various sectors including art, science, technology, engineering, and industry. This paper anchors itself in the innovative educational theories of Sci-Tech Arts, incorporating case studies from holographic art to provide an in-depth analysis and periodic review of the higher education teaching system for this major. The paper is divided into five sections: it starts by tracing the emergence and development of Sci-Tech Arts, analyzing the strengths and weaknesses of international talent cultivation models and innovative practices in this field. It then discusses the disciplinary value and role of Sci-Tech Arts from multiple perspectives including science, technology, art, education, culture, and society. The third section compares and analyzes teaching philosophies, faculty structures, curriculum settings, postgraduate cultivation aims, and career trajectories within the discipline, summarizing the main characteristics of its teaching model; the fourth chapter describes the innovative practice paths at the Holographic Arts Center at the Beijing Institute of Graphic Communication, integrating aspects of research, creation, communication, industry, and science popularization to propose a development framework for the development of Sci-Tech Arts. Finally, the paper forecasts future trends in Sci-Tech Arts and offers targeted strategies and recommendations.

Recently, mobile museums have gained popularity by transcending traditional geographical limitations and delivering cultural artifacts directly to communities. Utilizing innovative holographic technology, these museums offer an enhanced viewing experience that allows detailed, multi-angle examinations of exhibits without physical contact, significantly increasing exhibition interactivity and immersion. This study is dedicated to utilizing advanced holographic display technology combined with flexible mobile exhibition design to present a novel viewing experience that transcends traditional museum interactions. The study not only showcases the practical applications of this technology but also develops a systematic approach to holographic mobile museum design thinking. Furthermore, it delves into the potential applications of holography technology in future museum displays, particularly in providing rich, diverse interactive cultural experiences, leading innovations in museum presentation, and opening new chapters in the transmission and popularization of cultural heritage.

Chinese funeral products serve as vital carriers of cultural heritage, bearing the living's commemoration and mourning for the deceased, and possessing profound significance on historical, social, and cultural perspectives. Innovation in funeral products is showing new trends with the advancement of technology and the diversification of artistic expression. Holography, as an artistic medium, can replicate virtually everything with light, offering a strong sense of space and emotional impact, thus providing new possibilities and methods for innovative expression of objects. With societal progress, there is an increasing emphasis on the environmental sustainability and emotional value of funeral products. Based on market research on current funeral products, this study has completed a series of experimental holographic funeral products. It explores the potential integration of funeral products with hologram and provide new modes of expression for traditional funeral culture.

The extended Ptychography iterative engine (ePIE) has the advantages of fast convergence and high imaging accuracy, and is widely used in quantitative phase imaging. In the measurement process of complex optical components, the coherent diffraction imaging may get stuck in local optima and stop converging, resulting in incorrect reconstruction results. To increase the convergence speed, a hybrid algorithm method was proposed, which improves convergence speed and reduces computation time. The principle of this method is the mixed reconstruction of Ptychographic iterative engine (PIE) and ePIE. By adjusting the proportion of PIE and ePIE in iterative calculations, the probe and the object can be reconstructed to the global optimal value simultaneously. In the experiment, a random phase plate with a minimum structure size of 9.6×9.6 micrometers was measured. The experimental results demonstrate that compared to the ePIE algorithm, this algorithm can significantly improve convergence speed and converge to global optima. This article proposes an accelerated convergence method based on Ptychography, which provides a solution for high-precision and complex distributed object measurement.

Digital holographic microscopy can achieve automatic focusing of the object under test by simulating the diffraction propagation of the object wave through a computer. However, existing automatic focusing techniques require multiple iterative calculations of the reconstruction results of the object wave at different diffraction propagation distances, and then selecting the optimal reconstruction distance based on criteria such as sharpness measure and energy concentration. This process is time-consuming and the calculation accuracy depends on the iteration step size. This paper proposes a dot-matrix assisted focusing method for parallel phase-shifting digital holographic microscopy, which integrates the dot-matrix assisted focusing optical path structure into the parallel phase-shifting digital holographic microscope. This method allows the simultaneous acquisition of a polarization-modulated hologram and an auxiliary dot-matrix focusing image. By calculating the defocus amount based on the sharpness of the auxiliary dot-matrix, the optimal diffraction propagation distance of the object wave can be directly obtained, thereby reducing the computation time for diffraction reconstruction distance during hologram reconstruction and improving focusing accuracy. The measurement results of the micro-optical elements experiment verify the effectiveness and reliability of this method.

The Direct Binary Search (DBS) algorithm is recognized as an effective technique for the generation of binary holograms. DBS initiates with the generation of a stochastic binary hologram configuration, followed by a systematic scan in lexicographic order. During the scanning process, each pixel is individually addressed by inverting its value, subsequent to which the reconstruction Mean Square Error (MSE) is calculated. In instances where a diminution in the reconstruction error is observed, the inverted hologram value at the respective pixel is preserved. Conversely, if no reduction is detected, the pixel reverts to its original value. This iterative procedure continues until a complete scan of the hologram is achieved without any inversions being maintained. For high-resolution holograms, the Direct Binary Search (DBS) algorithm emerges as an impractical approach due to the enhanced computational load in each iteration, a consequence of the higher resolution. Moreover, the algorithm's search order predisposes it to settling on local optima rather than global optima. This propensity significantly diminishes the optical reconstruction quality of the computed binary holograms. In response to this problem, this paper proposes a Multi-Pixel Parallel Search (MPS) algorithm. The MPS algorithm adopts a multi-pixel addressing strategy in each iteration and utilizes the parallel computing method of GPU to accelerate the optical reconstruction calculation, the computation of binary hologram is significantly accelerated. The binary hologram calculated by DBS algorithm and MPS algorithm is reconstructed by optical experiments. The experimental results show that the binary hologram calculated by multi-pixel parallel search algorithm has better imaging quality than that by direct binary search algorithm.

Ancient mural relics in China are facing various erosive deteriorations, with surface and subsurface microcracks formed by internal and external factors being the primary origin. Therefore, the non-destructive detection, formation mechanism, and degradation trend analysis of microcracks are crucial in the restoration and preservation of ancient mural relics. To comprehend the growth process of mural cracks correctly, a digital holographic deformation detect system of large field of view (FOV) for in-situ detection of large murals was established. Combined with Gaussian 1σ criterion and histogram segmentation method, quantifiable three-dimensional distribution of microcrack structures can be obtained. According to the structure of defective mural, microscopic molecular dynamics model of the material particles and voids of the defective micro-nano layer in the mural was established to analyse the expansion mechanism and degradation mechanism of mural microcracks. Results indicate that the system can achieve three-dimensional characterization of elongated microcracks with extensive ranges. Molecular dynamics simulations reveal that temperature cycle affects stress distribution, leading to the fracture and recombination of atomic bonds. The stress reaching threshold partly triggers secondary fracture of dislocations, serving as the root cause of sub-crack formation. This suggests that the temperature of environment influence the expansion speed, trend, and direction of microcracks, consistent with the macroscopic evolution trend of pathological cracks. This study, integrating digital holography of large FOV with molecular dynamics simulation, provides a new reference for the study of the formation mechanism of ancient mural microcrack damage and the formulation of cultural heritage protection strategies.

Small-sized, highly sensitive pressure sensors are crucial in the field of turbomachinery application. In this paper, we present a l low-cost and user-friendly method for producing PDMS films with thicknesses about 10μm and the preparation method and process are described in detail. The PDMS films we fabricated have thicknesses of 11.5μm and 9.2μm. After the prepared FPI was cured and packaged, the characteristics of the sensor were studied theoretically and experimentally. The pressure sensitivity of the two groups with similar cavity length is 23.04nm/Mpa and 29.64nm/Mpa. This method provides a novel approach for the large-scale, low-cost production of 10-micron-level PDMS thin-film FPI.

The Brillouin scattering is one of the most important nonlinear effects in optical fibers. Measuring and characterizing the Brillouin gain spectrum is vital for any practical Brillouin applications. The Brillouin spectrum is commonly measured with very long optical fibers of at least a few hundred meters to several kilometers in length to ensure the generation of sufficient Brillouin gain. Not only the measuring system becomes cumbersome, but more importantly, what if the fiber under test cannot be made long under no circumstance. In this paper, we propose and demonstrate a specially designed Brillouin gain measuring system, based on an ultra-narrow linewidth pump laser with a heterodyne detection scheme, that is capable of characterizing the Brillouin spectra for optical fibers as short as a few meters in length. Single-mode optical fibers in 10-meter-long and 1-meter-long segments have been successfully measured and analyzed, while the minimal working length is expected to be shrinking down to a couple of meters even for large-mode-area fibers by theoretical prediction.

As the global population grows and climate change intensifies, sustainable water management and agricultural practices face serious challenges. This paper introduces an innovative distributed optical fiber sensor based on Optical Time Domain Reflectometry (OTDR) and employing superabsorbent polymers (SAPs) as humidity-sensitive materials. SAPs, known for their high water absorption and retention capacities, are utilized here not only to amend soil properties but also to facilitate precision in soil moisture and water retention monitoring. The sensor system comprises an OTDR device connected to fiber optics integrated with SAPs through a novel structural setup where each node contains a SAP humidity-sensitive materials formed after SAP solidifies and is tightly installed in the groove structure above the optical fiber. This design can detect humidity changes through the expansion or contraction caused by the SAP humidity-sensitive materials, thereby driving the pressing or releasing of the optical fiber, changing the bending of the optical fiber, and thus affecting the attenuation characteristics of the optical signal. These changes are precisely recorded by the OTDR, with different attenuation levels translated into moisture levels via a calibration chart. The sensor demonstrates high accuracy and spatial resolution in laboratory settings, achieving moisture monitoring with a resolution up to the meter-level, suitable for detailed field applications. Additionally, the system's real-time data transmission capabilities allow for rapid response to moisture changes, supporting real-time decision-making for precision irrigation. This technology shows potential in enhancing irrigation strategies, improving crop yields, and conserving water resources, indicating a promising avenue for broader application in precision agriculture.

Optical fiber access has the remarkable advantages of large communication capacity, low loss and good confidentiality performance, which can meet the high demand of modern network communication, and has become a mainstream way of broadband access. Starting from the working principle of optical fiber broadband access, this paper mainly analyzes the network architecture of optical fiber broadband network system and the performance parameters of each component unit, analyzes the existing problems in the application process of optical fiber broadband network construction, and obtains the implementation method of fault location and fault handling to solve the problem of broadband network failure.

Portable, accurate and low-cost fruit counting is critical for estimating early capsicum yields. On the basis of lightweight YOLOv8n deep learning model, an improved YOLOv8n model is proposed. Firstly, by integrating the Ghost module in the backbone and head, the number and complexity of model parameters can be effectively reduced. Then, by integrating the SKA attention mechanism in the C2f module, the model ability to capture spatial information is enhanced, and a good balance between computational efficiency and computational performance is realized. By replacing the Conv module with the ODConv module, the convolution kernel is dynamically adjusted, which effectively enhances the feature extraction ability of the model for the target region. By replacing the CIoU loss function with the EIoU loss function, the convergence speed in the model training process is accelerated and the model detection accuracy is improved. Finally, experiments are carried out based on the capsicum dataset under complex natural environment conditions to verify the generalization ability of the improved YOLOv8n model. Compared to the original YOLOv8s model, the improved Yolov8s model improves the precision, recall and mean average accuracy (mAP) by 2.4%, 4.8% and 3.9% respectively, with an average detection time of 5.69 milliseconds and a model parameter of only 5.3M. The experimental results show that the model proposed in this study has superior automatic counting performance, and can provide farmers with a fast, accurate and cost-effective tool for estimating early capsicum yield.

With the development of smart agriculture, automatic picking robots have become a research hotspot and achieving accurate recognition of fruits and low-cost deployment are the key technologies. A lightweight strawberry detection model YOLOv10-SD (strawberry detector) is proposed to address the problem that strawberry targets are small and susceptible to occlusion by branches and leaves. Firstly, StarNet is used to reconstruct the backbone network of YOLOv10n to reduce the number of parameters and achieve fast detection. Then, the Coordinate Attention (CA) module is introduced to enhance the accurate recognition of the region of interest or the position of important objects in the convolutional neural network, thereby improving the detection effect of strawberries under complex backgrounds. Finally, this model integrates Inner-IoU into CIoU as the loss function, accelerates the regression process of low IoU samples through a larger auxiliary bounding box, and is conducive to enhancing the detection ability of small target strawberries. Experiments are carried out on a strawberry dataset under a natural growth environment. The results show that compared with the current mainstream object detection algorithms YOLOv5n, YOLOv7-tiny, YOLOv8n, YOLOv9-c and YOLOv10n, YOLOv10-SD exhibits better comprehensive performance in terms of model parameter quantity, real-time performance and detection accuracy. The YOLOv10-SD model proposed in this study improves the ability to identify and locate strawberries in natural environments while ensuring lightweight, providing a technical reference for efficient smart strawberry picking.

The far-field intensity of lasers is an essential parameter for long-range laser emission, affecting the performance in laser countermeasures, laser imaging, and laser communication, etc. Predicting the laser distribution and power density before actual emission is therefore crucial. During long-distance air transmission, the far-field intensity distribution is influenced by atmospheric factors such as scattering and turbulence. The conventional calculation method utilizes separate visibility and turbulence testing equipment to abstract the current atmospheric conditions into a few parameters, followed by the application of physical formulas or empirical rules to estimate a result, which causes the loss of detailed atmospheric information. This paper presents a novel method for predicting intensity distribution. By performing analysis on the image acquired from the light path of the laser emission system, we can extract the atmospheric factors and predict the laser distribution using deep learning models, thereby avoiding the need for additional equipment and enabling quasi-real-time prediction. Simulation and experiments demonstrate that the new method achieves higher accuracy and has the potential to provide computation results for in-time decision.

Correspondence imaging (CI) realizes object reconstruction by conditionally averaging random patterns with a constant threshold which is generated from light intensities recorded by a single-pixel detector. However, object reconstruction in CI meets its challenge in complex media where the recorded light intensities fluctuate due to scaling factors existing in the optical channel. In addition, CI suffers from low quality because of conditional averaging among random patterns. In this paper, we report a varying-threshold-based CI that enables object reconstruction with high quality through complex media. To eliminate effect of dynamic scaling factors caused by complex media, varying thresholds are estimated by building an optimization model to optimize consistency between the estimations and light intensities. Then, the estimated varying thresholds can be utilized to binarize light intensities. In addition, to improve reconstruction quality, an optimization model is built by minimizing the L1 norm and total variation (TV) norm. We demonstrate the method using optical experiments. It is verified that the method can eliminate the effect of dynamic scaling factors and realize object reconstruction with high quality.

Defocus blur is one of the most commonly arising types of artifacts when taking photos. Blind defocus image deblurring is a classical ill-posed problem, as we have to estimate two variables with one observation. The blur in the defocus image is usually spatially varying which makes the estimation of clear image more challenging. In this paper, we propose an end-to-end blind defocus deblurring network (DefoGAN) by repeatedly incorporating an De-RES Block. DefoGAN network is based on the generate adversarial network, while utilizing feature pyramid network (FPN) for multi-scale feature extractor. The proposed DefoGAN is shown to improve the generated images’ quality and authenticity and achieve a balance of quality and speed than other compared methods. Ablation studies and evaluation metrics proved the effectiveness of the proposed method.

When random patterns are applied, correlation-based reconstruction is widely used in single-pixel imaging (SPI). The performance of the correlation algorithm still needs to improve in achieving high-quality object reconstruction, although some methods have been developed (e.g., Gerchberg-Saxton-like). Here, we present an approach to enhancing SPI performance by integrating the reweighted amplitude flow (RAF). The method optimizes the reconstruction process by weighting the measurement data adaptively to improve robustness and reconstruction accuracy. An efficient estimation of the object is first obtained through the weighted maximal correlation initialization. Subsequently, iterative updates refine the estimate using the reweighted gradient descent. This approach improves SPI performance, providing high-quality object reconstruction. The results demonstrate effectiveness of the RAF-enhanced SPI, showing its potential for the applications.

Currently, many underwater polarization imaging methods are not suitable for scenes containing targets with different polarization properties because they often assume that the target reflected light is highly depolarized or that the degree of polarization of the light is constant and it will cause the loss of some targets’ energy in the restored image, affecting the overall contrast of the image. In this paper, in order to solve this problem, the intensity, the degree and angle of polarization of backscatter light in background region are calculated to further calculate the transmittance in this region, and then the global distribution of backscatter light’s intensity and transmittance are obtained by the extrapolation method. Since the influence of the angle of polarization of backscatter light on the transmittance calculation is considered and the transmittance is obtained by the extrapolation method, the calculated transmittance is more accurate and can ignore the influence of the polarization properties of the targets. Experiments results show that the proposed method is suitable for scenes containing targets with different polarization properties. In addition, the advantage of the method is proved by comparing with another underwater polarization imaging technology and the value of EME is increased by 30.7%.

In recent years, many studies have successfully leveraged convolutional neural networks to improve the performance of In recent years, many studies have successfully leveraged convolutional neural networks for enhancing object detection performance in remote sensing images. However, the characteristics of optical remote sensing imaging pose several challenges, such as the small size of objects, which complicates feature extraction. Additionally, optical remote sensing images are often affected by complex background information and adverse weather conditions, which heighten the challenge of distinguishing objects from the background. To address these issues, this paper proposes an efficient small object detection model for optical remote sensing, called Multi-Scale Feature Fusion YOLO (MFSS-YOLO), to enhance detection accuracy and reduce the model's parameter count. MFSS-YOLO consists of two modules: the Replicated Hybrid Global Stem (RepHGStem) module and the Multi-Scale Feature Fusion (MFSS) module. The RepHGStem enhances the model's capabilities for contextual information capture and feature extraction by expanding the receptive field of the stem. The MFSS module further enhances the model's multi-scale feature extraction capability. Extensive experiments were conducted on two public small object datasets (AI-TODv2 and TinyPerson), and the results demonstrate that MFSS-YOLO achieves satisfactory performance in terms of both efficiency and accuracy, showing advantages in detection accuracy and model efficiency.

Due to the limitations of traditional single-modal object tracking methods under challenging conditions, RGBT single-object tracking has gained attention for its robustness. In this paper, we introduce a new RGBT object tracking model, SiamTFF, which utilizes a dual-modal feature interaction network to efficiently harness complementary features from both visible and thermal infrared information, thereby enhancing the model's ability to track objects under various challenging conditions, such as low-light or background clutter. Additionally, we have designed a multi-level cross-correlation operation structure, enabling the model to effectively track targets of various sizes. SiamTFF based on a siamese network framework, which balances speed and performance more effectively compared to models based on transformer and MDNet frameworks. We trained and tested SiamTFF on public datasets. The tracking results indicate that our model achieves high precision and success rates in real-world tracking scenarios while effectively balancing performance and speed. This demonstrates the effectiveness and generalization capabilities of SiamTFF, as well as its ability to handle various challenging conditions efficiently.

Adaptive Optics (AO) technology is an effective method for compensating atmospheric turbulence. However, inherent system delays restrict the correction performance, preventing real-time compensation. To overcome these limitations, forward prediction of atmospheric turbulence has become crucial. In this study, we propose a spatiotemporal prediction model based on the Frequency Graph Neural Network (FGNet) to address the challenge of accurately predicting random atmospheric turbulence. FGNet embeds Zernike coefficients, representing multiple frames of wavefronts, into a hypergraph, with each coefficient treated as a graph node. Meanwhile, we transform the temporal domain into the frequency domain, utilizing graph Fourier operators to capture the spatiotemporal features of the data. Compared to traditional AO methods and LSTM-based deep learning models, our proposed FGNet achieves superior prediction accuracy and stability on simulated data, offering an innovative and efficient solution for real-time atmospheric turbulence correction.

Pulse thermography is a reliable method for detecting defects in composites with advantages of full field, non-contact and easy operation. However, the size determination of subsurface defects with low width-to-depth ratios using pulsed thermography is a challenging task due to the blurred defect edges in infrared thermal images caused by lateral heat diffusion. The present study explores the ability of time derivative images of infrared thermography to predict defect size. 2D thermal conduction of pulse thermography based on the virtual heat source method was analyzed to compare the temperature profile and derivative profile of defects along the diameter direction, and a size prediction method based on the derivative profile is proposed. The feasibility of the proposed method was evaluated by measuring the dimensions of the flat-bottomed holes in the composite panels. The experimental results demonstrated that the size inversion method based on derivative images brings an improvement in measuring accuracy for defects. Additionally, the influences of cooling time, defect depth, defect size and thermal properties of the material on the measurement of defect dimension are also discussed.

Infrared thermography (IRT) is a reliable method for detecting defects in composites with advantages of full field, non-contact, easy operation and good visualization. Nevertheless, interpretation by experts is required to distinguish between defective and sound regions in the practical evaluation of defects, which limits the industrial applications of infrared thermography. In this study, a denoising diffusion probabilistic model (DDPM) framework named IRT-Diffusion is proposed to automatically segment defective regions in thermal images. IRT-Diffusion can reconstruct a defect segmentation image from a noisy image with standard Gaussian distribution by iteratively performing multiple denoising operations. Detection results from various traditional thermal signal processing methods are employed as input for the conditional noise predictor of IRT-Diffusion to generate more accurate defect segmentation results. The core innovation of this study is that the state-of-the-art generative model is first introduced and designed for defect identification in composites using infrared thermography. To assess the performance of IRT-Diffusion, experiments were conducted on several composites panels and compared with conditional variational autoencoder (CVAE) and conditional generative adversarial network (CGAN). The results demonstrate that the proposed method achieves superior quantitative metrics and effectively extracts defective regions.

In this paper, the generation of ultrafast pulsed lasers in the mid-infrared (mid-IR) region around 3μm is simulated by MATLAB and combined with Reinforcement Learning (RL) to train the mode-locking mechanism. Firstly, the mid-IR laser design and its generated ultrafast pulse characteristics mentioned in a paper are reproduced by numerical simulation. The simulation results verify the performance of the laser under specific parameter settings. Subsequently, we introduced a reinforcement learning algorithm to achieve an efficient mode-locked training process by continuously adjusting the laser parameters through an intelligent agent (agent). During the training process, the intelligent agent dynamically optimizes the laser parameters according to the feedback signals to achieve the optimal mode-locking state. The final experimental results show that the reinforcement learning algorithm has significant advantages in the laser mode-locking process, which not only improves the mode-locking efficiency, but also enhances the stability of the system. This study demonstrates that the combination of numerical simulation and reinforcement learning can effectively realize ultrafast pulsed laser mode-locking in the mid-infrared region, thus reflecting the richer application value of artificial intelligence in laser mode-locking and the potential for further research.

This study investigates the impact of Self Phase Modulation (SPM) on optical signals under varying input power levels, different modulation formats and fiber parameters in a 30km optical link, representative of a typical metropolitan backbone network. The increasing intensity of light signals in fiber networks introduces nonlinear optical effects and can significantly impair signal quality. Among these nonlinear phenomena, the Self Phase Modulation (SPM) effect induced by nonlinear refraction, results from the interaction between the optical pulses transmitted and the nonlinear response of the propagation medium. It causes phase and frequency distortions and produces a broadening of the signal spectrum. In this paper, through numerical simulations we analyse the evolution of optical pulses under the influence of SPM effect propagating in a single mode fiber. We consider a 30km optical link of a metropolitan backbone network with its real-parameters. The results reveal critical thresholds for signal degradation and demonstrate the interplay between SPM-induced spectral broadening and dispersion. We observe the broadening of the signal spectrum as the injected power increases independently of the modulation format. The findings provide valuable insights into optimizing metropolitan optical networks, ensuring robust performance under high data rates traffic conditions.

We propose an oblique layered metal-dielectric multilayer (OLMDM) with single slit to achieve nanoscale focusing and steering of light. It is theoretically demonstrated that it occurs with off-diagonal elements of permittivity tensor and asymmetry dispersion of OLMDM originated by oblique layered manner. For representative configuration of OLMDM combined with a 100nm slit, numerical simulations demonstrate that a minimum focus with FWHM 22nm (~λ/16) is obtained We also discuss those behaviors in OLMDM combined with different width slit.

Diamond nitrogen-vacancy (NV) center ensemble magnetometer is a novel solid quantum magnetometer with high sensitivity and high spatial resolution, which can be used at normal temperature and pressure. However, the low fluorescence collection efficiency of the magnetometer in confocal optical system hinders further improvement in sensitivity. To solve the problem, based on silica microspheres, we propose a method to enhance the excitation and collection efficiency of NV center ensemble in massive diamond. Experimental results show, compared with the NV diamond magnetometer without microspheres, the photon collection efficiency of NV diamond magnetometer coated with a layer of microspheres is increased by 110%. Thus, the sensitivity of the NV center magnetometer is increased by 4 times to 5.3μT/√Hz.

Abnormal blood supply is closely related to the occurrence and development of various diseases, such as cerebral infarction, varicose veins, and vascular tumors. Real-time quantitative detection of full-field blood flow changes in the lesion area is of great scientific and clinical significance for early diagnosis, intraoperative monitoring, and postoperative follow-up of related diseases. Existing blood flow imaging technologies such as magnetic resonance imaging (MRI), positron emission tomography (PET), laser doppler imaging (LDI), and optical coherence tomography (OCT), each possess unique advantages but also inherent constraints, hindering the simultaneous achievement of spatial and temporal resolution, real-time capability, and quantitative analysis. In this paper, we introduce a laser speckle blood flow imaging approach utilizing event cameras, which builds upon speckle metrology technology and capitalizes on the asynchronous response, high temporal resolution, and high dynamic range of event cameras. This method enables real-time blood flow imaging and velocity measurements. To verify the feasibility of the proposed method, we conducted experimental comparative research using simulated blood flow and mice as test subjects, benchmarking it against traditional laser speckle contrast imaging (LSCI) techniques.

Flexible screens consist of various layered structures, including a protective layer and a display layer. The display layer with limited deformation ability is expected to be in the neutral layer to minimize stress during folding. Ideally, the display layer with limited deformation capacity should reside in the strain neutral surface to minimize stress during folding. Current research predominantly focuses on measuring surface deformation, which fails to accurately capture the actual deformation of the display layer during bending. It is essential to study the deformation field of the display layer under bending conditions. This paper proposes a method for measuring the bending deformation of flexible screens using Digital Image Correlation (DIC). The proposed method combines displaying speckles on the screen with spraying speckles outside the screen, enabling concurrent measurement of both the display layer and protective layer of the flexible screen. The experimental results demonstrate that there exists a significant difference in the strain distribution between the display layer and the protective layer of the screen under various bending angles. This paper provides a new experimental tool for non-destructive testing, design parameters optimization, and manufacturing processes improvement for foldable screens.

This study applies chirped mirrors to dispersion compensation of visible green ultrafast lasers. We have designed a high reflectivity chirped mirror composed of TiO₂/SiO₂ dielectric films with alternating high and low refractive indices. The number of layers is 42, and the reflectivity is greater than 99.9%. The center wavelength is 522nm, the bandwidth is 20nm, the average group delay dispersion is -90fs², and the amplitude is ±10fs². The thinnest layer is 4.65nm and the thickest layer is 259.48nm. Afterwards, the refractive index of low refractive index materials can be calculated using precise equations to optimize group delay dispersion oscillations within ±2fs². The more accurate the refractive index value of low refractive index materials, the smaller the delay dispersion oscillation of the designed chirped mirror group, providing a new method for dispersion compensation of visible light band lasers.

Trending with the recent artificial intelligence development, deep reinforcement learning (DRL) algorithms have found vibrant applications for ultrafast mode-locked fiber lasers in practice. However, there has been lacking a comprehensive and detailed study regarding the algorithms’ performance in these laser systems up to date. In this paper, we present a comparative study of three popular DRL algorithms, the Double Deep Q Network (DDQN), Twin Delayed Deep Deterministic policy gradient (TD3), and Soft Actor-Critic (SAC), which are theoretically implemented in nonlinear-polarization-rotation-based mode-locked fiber laser systems. In our DRL performance simulations, the reward function is designed specifically to maximize the tendency towards mode-locking by the introduction of kurtosis in frequency domain. The simulated training processes indicate that the TD3 and SAC are outperforming DDQN in the multi-input context. It has been found that while DDQN has difficulty in handling multiple inputs and tends to have issues with convergence, the TD3 is the most stable and efficient, and the SAC is more advantageous in searching for various states. The acquired pulse evolution diagrams of the mode-locked states also have confirmed the feasibility of training ultrafast mode-locked fiber lasers with these DRL algorithms. Finally, an experimental ultrafast fiber laser system based on the active Ho:ZBLAN fiber and TD3 algorithm has been successfully demonstrated.