Point cloud is a kind of widely used 3D data, and semantic segmentation, as a key technology for 3D scene understanding, is increasingly in demand. In the past three years, point cloud semantic segmentation technology has been developing rapidly, and in order to show the progress in deep learning-based 3D point cloud semantic segmentation for indoor scenes, the latest research trends in the past three years are highlighted. Firstly, we introduce the commonly used datasets and evaluation indexes for point cloud semantic segmentation, then we classify the various point cloud semantic segmentation methods in the past three years, analyze and summarize the framework structure of the methods and their innovations according to different categories from the perspective of indirectly and directly dealing with point clouds, and compare and contrast the evaluation indexes of the various algorithms on several most commonly used indoor datasets, such as S3DIS, ScanNet, etc., such as the mIou indexes. metrics are compared and demonstrated. Finally, the current research status and existing problems of semantic segmentation techniques for point clouds are summarized and outlooked.

Semiconductor lasers have very strict requirements for operating temperature, and the threshold current, output optical power and wavelength of semiconductor lasers will change dramatically with the change of temperature. In order to ensure the normal operation of semiconductor laser, a temperature control circuit based on single chip microcomputer is designed, which includes temperature acquisition circuit composed of current source and instrument amplifier chip and thermoelectric cooler (TEC) drive circuit composed of power chip. Two pulse width modulation signals with different duty cycle are output by proportional-integral-differential (PID) algorithm to control TEC drive circuit. A method of reducing the pulse width modulation signal frequency to improve the output voltage modulation accuracy of TEC drive circuit is proposed to improve the circuit temperature control performance. The temperature control circuit of semiconductor laser based on STM32F103CBT6 is built to verify the feasibility of the experimental method. The experimental results show that when the ambient temperature is 20~50 ℃, the circuit can realize the stable control in the set temperature range of ±0.01 ℃.

High-power narrow line width fiber lasers have a wide scope of application, such as beam synthesis technology. The causes of the transverse mode instability (TMI) effect in high-power narrow linewidth fiber lasers and the methods for suppressing the TMI effect effectively are introduced, which can provide a reference for further increasing the output power of narrow line width fiber lasers.

Due to the stray light and electromagnetic interference in the environment, as well as the differences in the length and reflection characteristics of different paths, the FMCW liDAR system will be affected by multipath interference when receiving the target echo signal, and there is noise interference, which makes the accuracy of target ranging is low. Therefore, the FMCW LiDAR ranging method under the influence of coarse spatial resolution is proposed. According to the transmitting signal and echo signal of FMCWN laser radar system, IF signal is obtained. The EMD algorithm is used to decompose the IF signal, calculate the mutual relations between the IMF components of the signal, and determine and eliminate the IMF components with noise. The SFFT algorithm is used to refine the frequency spectrum of the IF signal after noise removal, obtain its estimate, determine the correction direction, correct the frequency estimate, and determine the distance of the measured target by using the corrected frequency of the IF signal. The experimental results show that the proposed method has good denoising effect, high spectrum estimation accuracy, high ranging accuracy and high ranging efficiency. The root mean square error of ranging is less than 0.04, and the longest time is only 0.23 ms.

Dark current, as an inherent noise in photodetectors, significantly affects the detection performance of dual-beam spectrophotometers. To mitigate this impact, an adaptive filtering dual-beam spectrophotometer is proposed. Firstly, a real-time dark current detection system is designed to capture the sample signal, reference signal, and dark current signal within each detection cycle. Secondly, an adaptive cascaded filtering algorithm called Recursive Least Squares and Normalized Least Mean Squares (RLS-NLMS) is introduced to address the issue of dark current elimination caused by temporal differences. The algorithm employs a three-level cascaded filtering model. The first level, local mean (LM), reduces the correlation between dark current and the detection signal. The second level utilizes a low-order RLS filter for signal filtering, while the third level employs a high-order NLMS filter to process the output signal from RLS. Experimental results demonstrate that the adaptive filtering dual-beam spectrophotometer achieves a relative error of only 1.46% in transmittance ratio measurement, reducing it by 66% compared to traditional methods. It exhibits excellent performance in dark current cancellation and holds significant practical value.

To solve the problems of high noise and low pixel of traditional industrial X-ray flat-panel detector, a 3 072×3 072-pixel X-ray flat-panel detector signal acquisition system has been designed by using an amorphous silicon thin-film transistor (TFT) flat-panel, a mature analog-to-digital converter(ADC) chip of ADI, a driver module of Himax, and a field-programmable gate array (FPGA). This project takes ZYNQ7020 as the core processor for circuit design, and realizes the comprehensive layout and wiring, simulation and debugging of FPGA. The system, under the action of the control signal of FPGA, implements the driver module to control the switch of the TFT, the analogue front-end (AFE) to collect the charge signal of 3 072 channels in parallel at the same time, and then the data is converted into the signal by denoising, amplifying, and AD conversion in a Low Voltage Differential signal (LVDS) interface serial output, and finally successfully obtained the acquisition signal, verified the function of the system, which can provide a design idea for the study of X-ray detectors.

In this paper, a method of gluing quality detection for curved workpiece based on image and point cloud is proposed, which solves the problem of low accuracy and poor robustness of image-based gluing quality detection. The method includes using the circular Mark point for rough positioning, introducing the improved iterative closest point algorithm with normal vector consistency constraint to complete the fine positioning, and extracting the glue skeleton information from the image into 3D glue trace points. The sampling points were obtained by equidistant and ordered sampling method to detect the quality parameters of the glue line. According to the normal constraint of sampling point and tangential constraint of glue trace, the sampling glue trace cross section model was obtained, and the point cloud was mapped to the cross section to get the glue trace cross section profile model. Experimental results have shown that the measured width error of the glue line is less than 0.35mm, and the thickness error is less than 0.25mm, which meets the quality evaluation requirements of glue lines in industrial scenarios.

To address the problem of poor detection performance of giant pandas in complex environments and the low efficiency of object detection models on resource-limited embedded devices, a lightweight giant panda detection model called GP-YOLOv5n is proposed. The model is based on YOLOv5n and improved by introducing a depth-separable neck network with attention, which enhances the detection accuracy and speed of targets in complex environments. Moreover, the model adopts Alpha-IoU in the bounding box regression loss function to improve the bounding box localization accuracy of targets. After training on a homemade giant panda dataset, the model is optimized for embedded devices and deployed on Jetson Nano. Experimental results show that the improved model achieves 97.8% and 73.6% in mAP50 and mAP50:95 metrics respectively, which are 2.7% and 9.2% higher than the original model. The detection speed of the model on embedded devices reaches 15.12 f/s, which can accurately and real-time detect the giant pandas in complex environments.

A vehicle detection algorithm based on deformable feature fusion is proposed to solve the problem of low vehicle detection accuracy in environment perception of unmanned driving system. We propose a 3D vehicle detection algorithm based on deformable feature fusion. Firstly, the convergence speed and performance were improved by the road scene enhancement algorithm. The ground point cloud is removed to reduce the interference of irrelevant point cloud. Then, a deformable feature fusion module was constructed to adaptively align the pose and position information between different modal data to improve the utilization efficiency of multi-modal data. The loss function was optimized, and the adversarial loss was added to judge the authenticity of vehicle motion, so as to improve the detection accuracy of the network for small targets. Finally, the best weight of the network model is obtained by training, and the KITTI data set is used for testing, which can achieve better vehicle recognition effect. The experimental results show that the average precision value is 83.26%, and the average detection time is 0.15 s. The algorithm can quickly and accurately identify the vehicle in the unmanned driving system.

This paper proposes a fingerprint detection system that integrates polarized imaging with spatial filtering. The system effectively extracts fingerprint information submerged in background noise, achieving high-quality imaging of fingerprint targets on phase objects. By using a focusing lens and a pinhole reflector, low-frequency background patterns of Fraunhofer diffraction are filtered out, capturing only high-frequency detail information. With the integration of polarized imaging technology, the polarization parameters and Polarization Degree of Linear Polarization (PDOLP) images of fingerprint images are obtained, resulting in clear fingerprint imaging. The results show that the addition of spatial filtering and polarized imaging technology improves the clarity of fingerprint images by approximately 6.8 times, highlighting details and ridges of fingerprint. Compared to intensity images, PDOLP images of fingerprint can significantly eliminate the influence of other interfering substances on phase objects, obtaining more complete and rich fingerprint detail information. This technology holds potential applications in fingerprint verification and criminal investigation fields.

Lane line detection, as the main research direction for safe driving of intelligent vehicles, can provide timely warning when the vehicle deviates from the lane, effectively alleviating traffic congestion and safety issues. However, conventional methods are easily affected by environmental factors such as light intensity and shadows, limiting their scope of use and causing significant detection errors. Therefore, a laser precise detection method for lane lines based on inverse perspective transformation is proposed. This method uses RS-LiDAR-16 LiDAR as the lane data acquisition device, uses inverse perspective transformation and top view spatial coordinate system to convert various laser point data, and uses the maximum and minimum inter class variance algorithm to find the optimal threshold of laser point reflection intensity, which serves as the basis for judging the surface and lane line data of the lane. The data of each point of the lane line is obtained through binary calculation, and these data are fitted into a line using the least squares fitting method, Finally, the lane line was detected. The experimental results show that the proposed method has high accuracy in lane line detection, and the inverse perspective transformation reduces the interference of the environment on the detection results.

In order to improve the accuracy of personnel intrusion detection in the work area, this paper proposes a method for personnel intrusion detection in high-voltage field work areas based on infrared grating technology. This method utilizes infrared grating technology to collect visual information of personnel images, and analyzes structured similarity features, combined with supervised comparative learning and backbone feature extraction methods, to achieve feature segmentation and reconstruction processing. At the same time, the edge bounding contour features of personnel are obtained through ambiguity analysis to achieve recognition of personnel intrusion behavior. After experimental verification, the results show that this method has high accuracy in intrusion detection, with an average false alarm rate (FPR), false alarm rate (FNR), and false alarm rate (FAR) of 2.69‰, 3.13‰, and 3.21‰, respectively, and a detection rate of 16.25 frames/s.

Workpiece anomaly detection is a key link in production. Due to the small number of abnormal samples and large randomness, supervised learning can not fully learn all types of anomalies, and there exists the problem of poor model stability. In order to solve the above problems, this paper studies an unsupervised workpiece anomaly detection algorithm based on reverse knowledge distillation, and uses the teacher model and student model designed by ResNet network structure as the backbone network. The teacher model truly extracts the image features, the student model reconstructs the image according to the prior knowledge, and adopts the reverse structure to expand the specificity of the abnormal condition. A memory module and a mask attention module are added to extract the multi-dimensional feature information of the sample to avoid missing the details of the image; after the memory module, a mask attention mechanism is added to integrate the multi-dimensional and multi-level features of the image. the accuracy of detection is further improved. The experimental results on two open industrial anomaly detection data sets show that the proposed algorithm is 5% higher than the general knowledge distillation algorithm AUC by 7%, and the effect of locating subtle anomalies is better.

In order to improve the effectiveness of object detection, a method for object detection in laser remote sensing images based on residual dense blocks is proposed. Firstly, design a convolutional neural network based on residual dense blocks. After designing the ReLU activation function and completing network training, based on the preliminary feature extraction results of noisy laser remote sensing images, use a single convolution to unfold the convolutional mapping process and extract potentially clean images. Then, through clustering processing, the saliency map of vehicle targets in the laser remote sensing image is obtained, and then the target information is detected using the established feature proportion relationship using the general law. The experimental results show that the application of this method effectively filters out noise in laser remote sensing images and accurately detects vehicle targets in laser remote sensing images. Compared to the three traditional methods, the minimum value of the mean error of the detection results of this method is only 0.015 6, indicating that this method effectively achieves the design expectations.

Photo-Response Non-Uniformity (PRNU) noise can be used as the fingerprint of the camera and source camera identification of digital images because of its uniqueness and stability. In order to improve the accuracy and efficiency of source camera identification, this paper proposes a PRNU noise extraction algorithm based on U-shaped Transformer deep network (Uformer). The network uses a Transformer block based on Locally-enhanced Window (LeWin), which can effectively extract local context information with low computational complexity. Secondly, the network uses a Multi-Scale Restoration Modulator in the form of multi-scale spatial deviation, which can adaptively adjust the multi-layer features of the Uformer decoder, so as to better extract the potential PRNU camera fingerprints in the image. The experimental results on the Dresden dataset show that the AUC values of the proposed algorithm at 128×128 pixels, 256×256 pixels and 512×512 pixels are 0.836 8, 0.925 0 and 0.972 0, respectively, and the Kappa values are 0.900 5, 0.744 7 and 0.473 7, respectively. They are better than the existing methods.

Compared with the traditional artificial neural network, pulse neural network has the advantages of hardware-friendliness and low energy consumption. Compared with the electric pulse neural network, the photon pulse neural network has the advantages of high speed, low energy consumption, large transmission capacity and strong anti-electromagnetic interference ability. Based on the semiconductor optical amplifier (SOA), a new neuron model of photon pulse—LIF (leakage integration and fire) model is designed in this paper, and the application of LIF model in digital logic is explored. The “XOR” logic function based on the model is realized, and an excitatory and inhibitory neuron is improved, LIF model, XOR logic and good output of excitatory and inhibitory neurons were obtained by optical communication simulation software.

In order to improve the path optimization effect and load balancing ability of industrial fiber optic communication network, the routing optimization method of industrial fiber optic communication network under 5G and digital twin is proposed, firstly, the industrial fiber optic communication network is modeled and analyzed with the combination of 5G technology and digital twin technology, and the network state data is obtained; secondly, the minimization of the maximum link utilization rate and the path latency is taken as the objective, and the routing optimization objective function of the industrial fiber optic communication network is established. Finally, the improved sparrow algorithm is used to solve the objective function to realize the optimization of industrial fiber optic communication network routing. The experimental results show that the maximum link utilization of the proposed method is within 20%, the maximum path delay is 0.27 s, and the number of dead nodes is 1/8 of all nodes, which verifies that the proposed method has a good path optimization effect and load balancing capability.

The traffic data in optical communication networks has the characteristics of large-scale and high dimensionality, and the inconsistency of data dimensions amplifies the differences between the data, resulting in unsatisfactory interpolation effects. Therefore, a continuous interpolation method for optical communication network traffic data based on improved transfer learning is proposed. The Box-Cox transformation method is used to standardize the traffic data and unify the data scales and dimensions. The convolutional neural network is improved using deep learning theory and VNet technology. By updating the network parameters, the continuous interpolation results are matched with the ideal data, obtaining the continuous interpolation results of the traffic data. Experimental results show that the signal-to-noise ratio of the proposed method is always higher than 27.83 dB, and the frequency-waveform distribution graph is most similar to the ideal data, with a coefficient of determination above 0.8, which can obtain high-quality interpolation results.

With the increase of network traffic, the complexity of dynamic routing calculation in fiber optic networks increases, consuming more joint resources. Therefore, a joint resource optimal allocation method of Dynamic routing in optical fiber networks is proposed.. Map network nodes in the form of a directed graph to calculate the spectrum utilization of routing transmission. Obtain Dynamic routing joint resources according to bandwidth, delay, and network delay jitter indicators. By selecting the spectrum service with the best continuity through the spectrum continuity of the transmission path, the service is segmented and merged based on spectrum sensing and greedy algorithms to achieve optimal joint resource allocation. Experimental results show that when the traffic load is 400 Erlang, the bandwidth Constrictively of the proposed method is only 10%, and the spectrum utilization rate is as high as 90%. In addition, after a dynamic resource transmission cycle, the proportion of data nodes in this method is low, indicating that this method has good stability.

An Iterative attention normalization flow (IANFlow) network is proposed to address the problem of insufficient feature fusion between network layers and lack of accurate localization and acquisition of high-frequency features, as well as the problem of uncertain mapping between low-light images and multiple normal-exposure images. The iterative attention module uses spatial and channel attention to localize the high-frequency feature regions of the input feature maps and then performs feature acquisition, which prompts the deeper feature maps to contain more high -frequency features through incremental hierarchical localization and fusion; the reversible normalization flow module learns the complex conditional distributions between low-light images and normal-exposure images as well as minimizes the negative log-likelihood (NLL) to establish the uncertainty in mappings between a low-light image and a reference image. one-to-many mapping. The peak signal-to-noise ratio (PSNR) of the IANFlow network is improved by 1.1 dB, 1.27 dB, and 2.14 dB when comparing the LLFlow network on each of the three datasets.

The overexposure of an image directly affects the reading effect of effective information in the image. Laser images have the characteristics of high energy and high brightness, which makes them prone to overexposure. In order to restore the information in overexposed laser images, an intelligent restoration method based on visual communication for overexposed excitation images is proposed. By simulating the overexposure process of laser images, degraded images are identified as the processing objects for image restoration. By comprehensively considering the histogram features and brightness features of laser images, determine the overexposed areas in the laser image. For overexposed areas, with the aim of improving visual communication, intelligent image restoration results are achieved through steps such as light compensation, color correction, and noise suppression. Through the effect test experiment, the conclusion is drawn: Compared with the traditional restoration method, the optimized design method obtained that the peak signal-to-noise ratio of the restored laser image reached 49.5 dB, the image restoration time was only 4.9 s, and the structural similarity reached 0.987, indicating that the method had a higher peak signal-to-noise ratio, the image was clearer and more similar to the original image, and effectively improved the image restoration efficiency.

In order to improve the accuracy and efficiency of dangerous target boundary crossing laser recognition, a research on dangerous target boundary crossing recognition based on laser image feature extraction is proposed. Using nonlinear direct transformation method to achieve scale invariant feature point matching in laser point cloud reflection image conversion; Design an anti sharpening dual mask image enhancement structure to optimize image enhancement processing; Perform Walsh transform and fusion on the reflection feature points of the point cloud, and use the BP neural network structure to identify dangerous targets that are out of bounds. The experimental results show that the proposed method enhances the detailed information of image features after application, does not have local exposure issues, has the lowest false detection rate and missed detection rate for dangerous targets exceeding the boundary, and has high recognition accuracy and efficiency.

The clarity of laser images collected in dark lighting environments will decrease, making it difficult to identify the image content and reducing its utilization value. In this context, in order to improve the visual image effect, a laser image optimization method based on visual image effect in dark lighting environment is studied. This method optimizes laser images from two methods. In the first part, NLM algorithm is used to denoise laser images, solving the problem of "infection" noise in laser images in dark lighting environments; The latter part combines the gamma correction method with the Retinex method, using the former to obtain the required illumination component for the latter and obtain the reflection component. Perform inverse operations on the optimized reflection component and illumination component, and recombine the brightness component to achieve optimal adjustment of laser image illumination in dark lighting environments. Visual image performance testing was conducted on the optimized laser image in a dark lighting environment. The results showed that the peak signal-to-noise ratio and average gradient increased after optimization compared to before. In addition, compared with traditional methods, the peak signal-to-noise ratio and average gradient in the study were relatively larger, with an average of 34.8 dB and 3.6, respectively. This proves the visual image effect of the studied method.

Infrared image enhancement can improve the details and contrast in the image, and make the target more clearly visible in the image, which is of great significance to the task of target detection and recognition in infrared images. Therefore, an infrared image enhancement method based on feature weighted fusion is proposed. Firstly, the contrast and edge details of infrared image are analyzed, and the denoising pretreatment of infrared image is carried out by combining spectral feature fusion method. After extracting the regional gradient features of the denoised infrared image, the extracted specific image features are weighted and fused, and the weighted and fused infrared image features are tone mapped by Retinex algorithm to realize the final infrared image enhancement processing. Taking mean square error, peak signal-to-noise ratio and structural similarity as experimental test indicators, the results show that the MSE, PSNR and SSIM of this method are 0.603%, 49.11 dB and 0.988 respectively. It is proved that this method can enhance the infrared image and has certain application value.

The presence of sample noise in color matching images can lead to significant errors between the color scheme and the standard color of the source image, resulting in poor color coordination comfort. To address the above issues, research is conducted on the interactive color matching design of image color interfaces for high-resolution laser scanning point clouds. In this study, high-resolution laser scanning technology was used to collect standard point cloud data as reference samples, and denoising was implemented for the point cloud data to reduce noise interference. Convert the RGB data of the point cloud into HSV data and establish a color matching model based on the Birkhoff meter values in the table. Combining the fruit fly algorithm and particle swarm optimization algorithm to avoid falling into local optima, the fruit fly particle swarm algorithm is used to obtain the optimal solution of the color matching model, and the optimal design scheme for interactive color matching of the image color interface is obtained. The results show that the Birkhoff meter value of color scheme 2 is 0.976, which is higher than the other five schemes. This indicates that color scheme 2 is the optimal solution for the color matching model, achieving the best color matching effect and making the color matching more coordinated and comfortable.

In order to obtain ideal hyperspectral images, a real-time optimization method for hyperspectral images based on denoising technology was designed to address the problems existing in current optimization methods. Firstly, analyze the current research progress of hyperspectral image optimization, identify the shortcomings of current methods, collect hyperspectral images, use adaptive threshold wavelet transform to denoise hyperspectral images, improve the quality of hyperspectral images, and then use Retinex theoretical model to enhance the denoised hyperspectral images, enrich the details of hyperspectral images. Finally, use convolutional neural networks for hyperspectral image classification, the test results show that the peak signal-to-noise ratio and average structural similarity of hyperspectral images optimized by this method are 31.18 and 0.981, which improves the quality of hyperspectral images and makes the classification accuracy of hyperspectral images exceed 92%. The optimization time of hyperspectral images is controlled within 4.5 seconds, which has significant advantages compared to other hyperspectral image optimization methods.

To address the problem that word vectors or character vectors are difficult to reconstruct the two-dimensional structure in mathematical expressions when using generative adversarial networks to generate images, the task of generating images with handwritten mathematical expressions is converted into a style conversion problem from printed mathematical expressions to handwritten mathematical expressions, and a self-constructed dataset with handwritten style categorization is used to train the style conversion model. In order to solve the problem of incomplete content, distorted details and low quality of images generated by CycleGAN network, a multi-scale discriminative cyclic consistency generative adversarial network MD-CycleGAN is designed, which introduces the CBAM attention mechanism to compensate for the loss of information in the downsampling link, introduces the ACON activation function instead of the ReLU activation function, and controls the network through adaptive learning nonlinearity degree of each layer. The experimental results show that the data enhancement method based on generative adversarial network in this paper can effectively reduce the degree of model overfitting. This study provides a new method for automatic recognition of handwritten mathematical expression images, which overcomes the data annotation problem and the model generalization problem, and has the potential for a wide range of applications, including the fields of mathematics education, scientific document processing, and mathematical search engines.

At present, the laser 3D image reconstruction method has some defects, such as low efficiency and long time. In order to obtain a better effect of laser 3D image reconstruction, a new high-precision and fast reconstruction method of laser 3D image is proposed. This method uses nonlinear improved method to calculate the position and spatial coordinate points of 3D camera, and makes its error distribution uniform. At the same time, the sparse matrix method is used to automatically detect the laser point cloud, remove noise and obstacles, and optimize the point cloud data, so as to get a more accurate laser three-dimensional image. The first-order partial derivative of laser three-dimensional image is calculated by differential approximation strategy, and the approximate edge position in the image is preliminarily determined. The edge features of three-dimensional images are extracted by ant colony algorithm, and the reconstruction coordinate points of edge features of three-dimensional laser images are calculated. Finally, the reconstruction of three-dimensional laser images is realized. The experimental results show that this method can extract the edge contour of laser 3D image with high accuracy, and the reconstruction accuracy is ninety-seven percent, and the reconstruction time is 35 ms.

This study investigated the manufacturing of micro-holes in Inconel 718 material using high-temperature chemical-assisted nanosecond laser machining. The study examined the effects of various pulse numbers on the diameter, depth, and wall characteristics of micro-holes in both ambient air and high-temperature chemical environments. Additionally, the study compared the morphological differences of micro-holes under ambient air and high-temperature chemical conditions and thoroughly investigated the material removal process involved in fabricating micro-holes using high-temperature chemical-assisted nanosecond laser machining. The experimental results demonstrated that using lower pulse frequencies had a beneficial impact on enhancing the morphology of micro-holes in high-temperature chemical environments. Finally, the study investigated the influence of pulse delay on the laser machining of micro-holes with high-temperature chemical etching assistance and analyzed the patterns of pulse delay in micro-hole formation. The results show that when the delay time is 1 s, the total processing time is 15 s, the roundness of the hole improves to 0.981, the hole wall becomes smoother, and the heat-affected zone decreases accordingly. Increasing a certain pulse delay can achieve high -quality micropores without recast layer, with a smaller heat-affected zone and better roundness.

It is very difficult to laser cladding copper-tungsten composite coating on pure copper surface, and the Marangoni effect plays an important role in the molten pool formed by laser cladding, so it is necessary to study the influence of the laser power scanning speed and the change of spot radius on the Marangoni effect. Therefore, in this study, COMSOL software was used to conduct numerical simulation of copper-tungsten composite coating on the surface of pure copper, and combined with convective heat transfer, the changes of Marangoni effect under different laser power, scanning speed and spot radius were analyzed. The results show that the smaller the cladding power, scanning speed and spot radius, the larger the Marangoni effect is. Therefore, the appropriate increase of laser power scanning speed and spot radius will weaken the Marangoni effect, so as to provide guidance for obtaining more excellent copper -tungsten composite cladding layer.

Laser shock strengthening can enhance the surface strength of aircraft engines and enhance material toughness. However, aviation engines are in a complex working environment, and their fatigue life is influenced by multiple factors. Especially the significant changes in material performance parameters after laser shock strengthening, which increases the difficulty of measuring the fatigue life of aviation engines. Therefore, a precise measurement method for the fatigue life of aviation engines after laser shock strengthening is proposed. Firstly, analyze the working principle of laser shock strengthening for aircraft engines, and establish a fatigue damage function for aircraft engines based on fatigue damage theory and Von-Mises yield criterion. Then, combined with the fatigue damage function and grey correlation analysis method, weight analysis was conducted on multiple factors that affect the fatigue life of the aircraft engine to obtain the fatigue index of the aircraft engine, thereby achieving accurate measurement of the fatigue life of the aircraft engine. The experimental results show that the proposed method can achieve precise monitoring of the fatigue life degradation trajectory of aircraft engines, and can accurately analyze the distribution of engine surface fatigue life, indicating that the method's fatigue life measurement results are more accurate.

Theoretical weight is the mainstream weight calculation method in current steel trade, thus a fast and accurate AI based rebar counting system has become a critical research issue. A design scheme of visual rebar counting system based on cloud-edge collaboration is proposed. The system uses a mobile terminal to capture rebar images, performs preprocessing including background removal and size adjustment, and uploads them to the cloud; The cloud uses a detection model based on an improved YOLOv5 to detect and count the rebars in the image, and feeds back the counting results to the mobile terminal. According to the test results in the SPDC, the counting accuracy of the proposed system can reach 99.85% after manual correction, which is higher than the average accuracy of manual counting. The time cost of intelligent counting per 1 000 rebars is 26.50 seconds after manual correction, significantly lower than the average time of manual counting.

During the milling process of a laser cutting machine, the surface may appear uneven or wavy due to workpiece deformation, with local areas of varying heights, resulting in the floating axis of the machine deviating from the actual motion trajectory and affecting the quality of workpiece milling processing. Therefore, a high-precision control technology for high-speed milling laser cutting machine tool CNC machining is proposed. Using kinematic theory and matrix transformation functions to establish a mathematical model of laser cutting machine tool motion, a high-speed milling control model of the laser cutting machine tool is constructed using PLC controllers, servo drivers, and motors. Negative feedback control algorithms and displacement error functions are used to control the floating axis motion of the machine tool milling process. That is, when the floating axis performs upward or downward floating motion, if the floating axis motion trajectory deviates from the preset trajectory, Adjust the parameters of the PLC controller to drive the motor to control the operation, causing it to move in reverse until it falls on the preset trajectory, achieving high-precision control of milling processing. The experimental results show that the milling quality of the proposed technology workpiece is high, and the maximum control errors of the upper and lower floating points during high-speed milling are 6 mm and 6 mm, respectively. The response time of milling control is 6.8 ms.

During the laser welding of the aluminum shell of the lithium-ion battery for electric vehicles, due to the high temperature, the gas in the welding area is easy to be converted into steam, thus forming pores. For this reason, a method of controlling porosity in laser welding of lithium ion aluminum shell of electric vehicle is proposed. The influence of laser welding process parameters (laser welding line energy, pulse frequency, and laser duty cycle) on the weld porosity of the lithium-ion aluminum shell of electric vehicles was analyzed. The approximate function of the model was obtained using the multi-output Gaussian process to build the parameter model for controlling porosity in laser welding of the aluminum shell of lithium-ion batteries for electric vehicles. The model was solved using the improved Ant colony optimization algorithms based on the direct learning mechanism to obtain the optimal solution for welding process parameters and achieve control over welding porosity. The experimental results show that the pulse frequency of the proposed method is 47 Hz at 350 s, and the operation time is only 20.97 s, which shows that the proposed method has good control effect and stability, and can realize the control of pores in the laser welding of lithium ion aluminum shell of electric vehicles, with high efficiency.

The high dimensionality of hyperspectral LiDAR data in the spectral dimension, which includes a large number of bands or frequency bands, it is easy to overlook useful information in the video spectral band, resulting in poor signal sorting performance of hyperspectral LiDAR. Therefore, a study on hyperspectral LiDAR signal sorting based on improved random forest is proposed. Firstly, the variational modal decomposition algorithm is used to denoise the noisy signal of hyperspectral lidar; Then, a long and short term memory neural network algorithm is used to extract features from the denoised hyperspectral LiDAR signal, and a self coding neural network is used to reconstruct the extracted features to obtain the reconstructed radar signal features; Finally, the random forest algorithm is used to complete signal sorting based on the characteristics of hyperspectral LiDAR signals. The experimental results show that the SNR of the proposed method is 30.648 dB, and the RMSE is 0.149 8. The predicted sorting category is almost consistent with the actual sorting category, and the analysis time does not exceed 5 s, indicating that the proposed method has good sorting performance and practicality.

Millimeter wave radar is a commonly used non-contact ranging technology. Due to environmental factors and various errors in the measurement process, the ranging results may have certain errors. Studying error compensation methods can effectively improve the ranging accuracy of millimeter wave radar, thereby obtaining more accurate distance information of target objects. Therefore, a high-precision millimeter wave radar ranging signal error compensation method based on machine learning is proposed. The noise in the radar ranging signal is removed through a Gaussian filter, and the signal denoising process is completed. The distance and angle information of the target object is measured using simulated insertion pulse counting method and four quadrant spot positioning method. The particle swarm optimization method is optimized through adaptive inertia weight and convergence factor, and the optimized particle swarm algorithm is used to improve the BP neural network, by inputting the measured distance and angle information into an improved BP neural network for training, the compensated radar ranging signal can be obtained. The experimental results show that the signal processing effect of this method is good, and the azimuth and elevation errors of the compensated millimeter wave radar ranging signal are close to 0, and the signal smoothness is high.

To achieve better strengthening effects and overcome issues such as uneven surface morphology after strengthening, a mathematical model for mechanical effects of laser shock strengthening with different wavelengths is designed. Using the Johnson Cook model to construct a material constitutive model of metal components, the basic material model is loaded with laser shock wave pressure, and based on the principle of laser shock strengthening, a finite element model of component laser shock strengthening at different wavelengths is constructed. Based on the constructed finite element model, this study explores the relationship between the dynamic yield strength of materials and the Hugoniu elastic limit of component materials, constructs a theoretical model between strain and peak pressure of shock waves, and represents the residual compressive stress generated by laser shock strengthening according to this model. The construction of a mathematical model for the mechanical effects of laser shock strengthening at different wavelengths is completed. The test results show that after strengthening based on this model, the surface morphology of the experimental components is generally uniform, with small pit depths. The displacement range of the component material surface on the straight line passing through the center of the light spot is between [-0.4, 0.5], indicating that the design model has good performance and practicality.

To control mobile robots to successfully avoid static and dynamic obstacles and safely and quickly run to their destination, a multi laser sensor fusion planning technology for obstacle avoidance path of mobile robots is studied. This technology installs two LiDAR sensors on the robot to sense the obstacle information in the environment where the robot is located. The obstacle positioning method based on multi laser sensor fusion weights the perception information of the two LiDAR sensors and extracts static obstacle position data. After that, the robot path planning method based on dynamic obstacle avoidance is used, and the improved A * algorithm is combined with static obstacle position data, Plan the global obstacle avoidance path for robots; Considering the possibility of dynamic obstacles appearing in the global obstacle avoidance path, the artificial potential field method is used to calculate the repulsive force, gravitational force, combined force, and direction of the dynamic obstacles and the destination on the robot, and adjust the direction of the robot's dynamic obstacle avoidance in the global obstacle avoidance path. In the experiment, the deviation of obstacle positioning under the method proposed in this article is within 0.1 m. This technology has qualified obstacle avoidance ability for static and dynamic obstacles. When running to the destination, it collides with the obstacle 0 times, and the planned path and running time are relatively short, with 158 m and 9.2 min, respectively, indicating high efficiency.