Abstract:Optical wireless communication is an important means of modern communication, with rich spectrum resources, antielectromagnetic interference and other advantages, can be used as an important supplement to traditional radio frequency technology, which is expected to provide an important technical driving force for future industrial manufacturing. This paper firstly introduces the relevant applications of optical wireless communication under the background of industrial Internet, and summarizes the research status at home and abroad in recent years, summarizes the typical research progress, expounds the channel model and key technologies of optical wireless communication system, and introduces the RF/ visible light heterogeneous technology used to ensure the reliable transmission of the uplink and downlink of the communication system. Finally, the paper summarizes the current problems faced by optical wireless communication in the industrial Internet, and looks forward to the future development trend, which can provide reference for the future research and development of optical wireless in this field.

Abstract:In order to improve the performance upper bound and generalization ability of current speech enhancement methods based on masking and spectrum mapping, a collaborative monaural speech enhancement method based on the learning framework of combined complex spectrum and masking is proposed. An interactive cooperative learning unit (ICU) is designed in the codec part to monitor the interactive speech information flow and provide an effective potential feature space. In the middle layer, a multi-scale fusion Transformer is designed to extract multi-scale details in the spatial-channel dimension with a small number of parameters for fusion output, at the meanwhile, modeling the voice sub-band and full band information. Experiments on large and small data sets and 115 noise environments show that the proposed method only uses 0. 57 M parameters to obtain better subjective and objective indicators than most advanced and related methods, which has good robustness and effectiveness.

Abstract:Precise point positioning (PPP) technology has been widely used in many fields because of its simple operation and high positioning accuracy. Aiming at the observation noise and multipath effect that may be caused by the change of surrounding environment, the traditional filtering algorithm cannot solve the problem of precision decline caused by it, this paper proposes a strong tracking adaptive Kalman filtering (SAKF) algorithm. The fading factor is introduced to adjust the prediction error value, and the measurement noise covariance is reconstructed by IGGⅢ function method, to achieve realize PPP solution. The experimental results show that the positioning accuracy of SAKF is improved by about 20% compared with the traditional algorithm in static solution, and it is improved by about 55% ~ 60% in quasi-dynamic solution, and it has better convergence stability.

Abstract:The identification of partial discharge type of switchgear has important guiding significance for understanding the insulation state and timely maintenance. The key to partial discharge type identification is to extract the characteristics of the partial discharge signal. A feature extraction method for partial discharge ultrasonic signals combining Choi-Williams distribution and permutation entropy is proposed, the time-frequency characteristics of partial discharge ultrasonic signals are obtained by using Choi-Williams distribution, the permutational entropy of partial discharge ultrasonic signals is solved, the complexity feature quantity of signal time series is obtained, the time domain and complexity features are composed into feature vectors, and the BP neural network optimized by particle swarm optimization is used to classify and identify discharge signals. The measured data analysis shows that the accuracy of the method for the identification of discharge type reaches 96. 67%, which is 11. 67% and 1. 67% higher than the traditional fractal and timefrequency analysis methods, respectively.

Abstract:To achieve the detection of surface crack defects on aluminum plates, a surface crack defect model is established based on COMSOL and the interaction between laser ultrasound and defects is analyzed. To deal with the problem of weak reflection signals and poor signal-to-noise ratio in the propagation of laser ultrasound inside materials, a signal multiple averaging combined with adjacent three-point difference processing method. Three cracks on the surface of 5 mm thick aluminum plate were detected using a laser ultrasonic visualization inspection system. Using multiple signals averaging can increase the signal to noise ratio (SNR) and enhance the damage echo in the maximum amplitude map. Extracting the peak and peak values of the signals at each scanning point in the target area reconstructs the three-dimensional maximum amplitude map. The ultrasonic signals in the horizontal and vertical directions are processed using an adjacent three-point difference processing method, providing better defect visibility. The results show that there is a significant interaction between surface waves ( R) and surface defects within a depth of 1 mm. The laser ultrasonic visualization inspection technology can quickly detect cracks and defects on the surface of aluminum plates, and can display the position and size of surface cracks in three dimensions. The multiple averaging and adjacent three-point difference processing method used can accurately characterize defects above 0. 5 mm, which will have extremely broad application value in industrial non-destructive testing and evaluation.

Abstract:In order to solve the problem of replacing TFT screens of different specifications when testing UHD video processing motherboards, and further shorten the test time, a system structure is proposed that combines the scale conversion and field frequency reduction of FPGA devices with DDR3 SDRAM memory chips to normalize video signals with different resolutions and field frequencies into HD video signals. The system uses 4 K@ 60 Hz UHD video as the input signal and sends it to the video data reading and writing module composed of DDR3 controlled by FPGA to realize cross-domain transmission, down-scale conversion processing and data connection; continuous output the HD video signal after down-scale conversion and field frequency reduction. After comparative experiments, compared with the storage structure of multi-channel FIFO plus DDR3, the storage resources consumed are reduced by 352 256 bits, and the conversion process time is reduced by 6. 761 μs. The results show that this system is more suitable for the testing requirements of video processing mainboards on production lines.

Abstract:The sensor structure and measurement circuit of electronic analytical balances are complex and cannot accurately calculate the transfer function of the system. In order to estimate the transfer function of the system and improve the signal-to-noise ratio of the measurement data, the order of the transfer function of the system is estimated by the method of Laplace transform, the equation of state of the system under steady state is derived, and the method of estimating the system parameters by autoregressive method and combining with Kalman filtering is proposed to filter the measurement data. Through the experimental estimation of equation parameters and noise intensity from offline data, the filtered data verify the smoothness of the process, and the significant level is as low as 0. 001. Compared with the commonly used sliding window filtering method, the smoothness and stability of the new method are significantly improved, the measurement standard deviation is 30% of the original method, the linearity can reach 6. 7×10 -5 , and the response time is as low as 10%. The experimental data of four samples verify the feasibility and effectiveness of the proposed method.

Abstract:Flexible hybrid circuit (FHC) is a composite circuit formed by integrating rigid electronic components onto a flexible substrate with printed electronics. In order to meet the increasingly complex application scenarios, FHC needs to integrate more rigid functional electronic components, which will lead to a sharp decrease in its flexibility. In order to alleviate the contradiction between high integration and high flexibility of FHC, this article starts from the perspective of structural improvement, which uses a hollow snakeshaped wire island-bridge structure (HS-SWI-BS) to carry out the optimization approach of FHC. The effectiveness and feasibility of this method are verified by three-dimensional finite element simulation experiments. The results show that the hollow snake-shaped wire island-bridge structure can improve the flexibility of FHC, and the increase of the flexibility of the model can reach 260%. The application and testing of flexible percutaneous electrical stimulation circuit have proved that this method has good practicality. The research has a reference role in improving the flexibility of FHC.

Abstract:In order to solve the problem of poor recognition effect of the plum blossom needle of the water meter in the process of factory calibration when it rotates at a high speed close to half of the acquisition frequency, a counting method of the plum blossom needle of the water meter based on the adaptive peak picking method is proposed. Firstly, the image is preprocessed and converted into a binary image. Then, the background XOR method is used to perform XOR operations on the current frame image and the starting frame image to obtain the motion trajectory of the plum blossom needle, and the change in the proportion of white pixels is calculated during the rotation process of the plum blossom needle. Finally, the number of rotating teeth is counted through the adaptive peak picking algorithm. The experimental results show that the algorithm overcomes the interference of false wave peaks on the peak picking algorithm. Compared with traditional statistical methods, this method has a significant recognition effect on the number of rotating teeth of plum blossom needles, with an error of no more than 1%.

Abstract:Appearance quality is one of the national mandatory verification for domestic gas meters (DGM). In view of the lack of defect samples in the appearance quality verification of DGM, which makes the detection method based on supervised learning difficult to generalize to the actual application scenario. This paper studies the unsupervised detection method of DGM appearance defects. EfficientFormerV2-l, the improved Vision Transformer (ViT), is introduced to extract normal sample features, fuse the bottom and highlevel feature maps, and map the normal features to the standard Gaussian distribution using two-dimensional normalizing flow called FastFlow. The appearance defects are scattered outside the distribution so that the abnormal score is higher than the normal sample. By setting an adaptive threshold, the DGM appearance defects are identified and located. The experiment collects DGM normal samples, real defect samples, synthetic defect samples as data sets and optimizes the detection model parameters. The optimized detection model achieves 99. 77% AUROC at image level indicators, 96. 3% AUPRO at pixel level indicators, and can detect more than 4 DGM images per second, indicating that the method in this paper can accurately and efficiently identify and locate DGM appearance defects.

Abstract:In the context of wireless sensor networks (WSNs), prone to internal node attacks, this study advances an intrusion detection approach underpinned by evolutionary game theory. The attack-defense confrontation of sensor networks is mapped into the game process, and the attack-defense game model between malicious nodes and cluster head nodes is established. The traditional replication dynamic equation is improved, so that the cluster head node takes the historical strategies of other nodes in the evolutionary game process to predict the attack strategy of malicious nodes. At the same time, the improved replication dynamic equation is applied to the intrusion detection algorithm to improve the response time of the algorithm. Experiments show that compared with the replication dynamic equation of the traditional method, the evolutionary game can quickly reach equilibrium by using this algorithm, and the convergence speed is 80% higher than that of the traditional method, which ensures the network security and avoids the consumption of sensor network detection energy.

Abstract:The high-precision measurement of the six dimensional acceleration sensor can effectively improve the control effect of the chassis anti-rollover control system, but the inter-dimensional coupling of parallel elastic elements can bring nonlinear errors to the sensor. The use of extreme learning machine algorithm for calibration and decoupling can effectively improve the measurement accuracy of the sensor. However, the traditional extreme learning machine nonlinear decoupling algorithm has low accuracy. The use of the sparrow search algorithm can obtain the optimal initial weights and thresholds of the extreme learning machine. At the same time, the maximum between-class variance method is integrated into the sparrow algorithm optimized extreme learning machine, which can explore the inherent coupling relationship of the six dimensional acceleration sensor. By converting the traditional black-box extreme learning machine model into a gray-box model, a decoupling algorithm for sparrow search optimization gray box extreme learning machine (SSAGB-ELM) is proposed. Through experimental verification, the nonlinear decoupling accuracy of the parallel six dimensional acceleration sensor using this algorithm is significantly improved, with a maximum error of 0. 023% for class I errors and 0. 046% for class II errors. The decoupling time is 1. 095 seconds, which can effectively solve the nonlinear coupling problem of six dimensional acceleration sensors.

Abstract:To solve the problems of low real-time performance and easy environmental impact in soft fault detection of redundant strapdown inertial navigation system (SINS), an APV/ FASPRT algorithm is proposed. Firstly, the parity space is constructed according to the hardware redundancy configuration, SPRT algorithm is implemented for the parity residual, and the fading factor and periodic reset are introduced to improve the tracking speed of the current residual information. Secondly, APV algorithm is used to detect the fault end time to reset the fading SPRT and provide sensitive axis information. In order to enhance the stability of fault detection, a threshold determination method based on the admissibility fault is proposed for the four and six gyro redundancy configurations commonly used in engineering. The simulation results show that compared with GLT, SPRT and APV methods, the average detection delay of the proposed algorithm is reduced by 50. 59%, 70. 21% and 2. 32%, and the average false alarm rate is reduced by 69. 31%, 99. 33% and 64. 77%, respectively. The real time of soft fault detection is enhanced and the false alarm rate is reduced in normal running time.

Abstract:In order to identify non-contact measurement of micro-hole diameter and depth, this paper proposes a measurement method utilizing eddy current effect. The measuring coil, composed of a single-layer multi-turn coil, traverses evenly over the target body during the micro-hole measurement process. The time variation of the coil’ s inductance correlates directly with the micro-hole’ s diameter. A linear relationship exists between the peak value of the coil’s inductance and the square of the micro-hole’s diameter, where the slope of this linear relationship is proportional to the depth of the micro-hole. According to the equivalent eddy current ring model, a mathematical model was established between the coil’s inductance value and the micro-hole’s diameter and depth. The influence law of the micro-hole diameter and depth on the coil’ s inductance value during the measurement process was analyzed by COMSOL finite element simulation. The simulation results were consistent with the analysis of the equivalent eddy current ring model. The eddy current measurement system for micro-hole diameter and depth was established, achieving measurements for diameters of 1. 5~ 5 mm and depths of 0. 1~ 0. 5 mm. When the micro-hole diameter is larger than 3 mm and depth greater than 0. 3 mm, the relative error in the diameter measurement stays within ±2%. The resolution of the depth measurement is 0. 01 mm. When the diameter of the micro-hole exceeds 2. 5 mm, the measurement error of depth is less than 0. 02 mm.

Abstract:In order to solve the problem of low node positioning accuracy caused by complexity of underwater wireless sensor network environment and node dynamics, a node movement prediction and positioning algorithm based on dynamic Bayes LS-SVM was proposed in this paper. In this algorithm, the distance and hop matrix from the beacon node to all beacon nodes within the communication radius are used as the training set, and the normalization process is carried out. The Bayesian LS-SVM model was constructed using Bayesian evidence framework, and the hop number vector between the unknown node and the beacon node was taken as the test set to determine the distance between the node and the beacon node. Then the equation of the distance matrix between the node and the beacon node was established and the maximum likelihood estimation method was used to estimate the coordinates of the unknown node. Finally, all unknown nodes were located by iterative method, and the adaptive increment and subtraction algorithm was used to dynamically adjust the model parameters and prediction model to adapt to the dynamic changes of data. The experimental results show that the average positioning error of the algorithm is reduced by 24. 77%, 22. 25%, 3. 1%, and 6. 5% compared with the SLMP algorithm, RTLC algorithm, NDSMP algorithm, and MPL algorithm under the same node density, effectively realizing underwater positioning. Dynamic positioning of unknown nodes.

Abstract:Aiming at the problem of serious distortion of inclination calculation caused by vibration interference of gravity acceleration of drilling tool in MWD, a method of gravity acceleration extraction of drilling tool based on federal adaptive unscented Kalman filter (FAUKF) is proposed. Firstly, a fusion extraction framework of federated gravity information without reset structure is established, and the recursive gravity value based on gyroscope data is selected as the common reference value of federated filtering. The combined gravity acceleration observed by decoupling geomagnetic data is used as sub-filter 1, and the combined gravity acceleration observed by accelerometer data is used as sub-filter 2. Then, the untracked Kalman filter (UKF) algorithm is carried out on the gravity state of the sub-filter. During this period, the vibration resistance of the gravity value observed according to the geomagnetic data is better than that of the accelerometer. The antivibration factor corresponding to the sub-filter 2 is designed. The antimagnetic factor corresponding to the sub-filter 1 is designed to improve the performance of the sub-filter, then the adaptive windowed factor is used to determine the windowed value in the windowed estimation rule and adjust the new information covariance. The real-time estimation of the sub-filter measurement noise covariance is achieved by estimating the new information covariance, and the accuracy of the untracked Kalman filter algorithm is improved. Then reliable local adaptive estimates of gravity information are obtained. Finally, the global estimation of gravity information is obtained by federal information fusion. The results of simulation drilling and real drilling experiments show that the inclination of FAUKF algorithm is reduced by ±1. 9° compared with FKF algorithm, and the inclination of FAUKF algorithm can be controlled within ±1. 2°. This method can effectively extract the gravity acceleration of drilling tools in coal mine, improve the measurement accuracy of inclination, and is an effective method to obtain reliable inclination.

Abstract:The high proportion and large-scale variation of small targets with defects caused by the complex structure of metal gear end faces have led to low detection accuracy, making it difficult to meet the real-time online detection needs of enterprises. In this paper, we propose a metal gear end face defect detection method based on an adaptive multi-scale feature fusion network (YOLO-Gear) using the YOLOv5s network. Firstly, we establish a gear end face defect detection test platform and create a gear end face defect dataset. Then, we introduce the adaptive convolutional block attention module (CBAM-C3) which combines channel attention module ( CAM) and spatial attention module (SAM) to enhance the adaptive feature learning and extraction for small target defects in metal gears, effectively improving the detection accuracy of the model for small target defects. Finally, we propose the bidirectional feature pyramid network (BiFPN), which repetitively weights and fuses features from different scales, thereby improving the model’s ability to detect defects at multiple scales. Experimental results demonstrate that the YOLO-Gear model achieves an average precision of 99. 2%, an F1 score of 0. 99, and an FPS value of 33 on the gear end face defect test set. Compared to other deep learning models, the proposed YOLO-Gear model in this paper improves both detection accuracy and efficiency, meeting the real-time online detection needs of enterprises.

Abstract:At present, the information fusion method of multi-sensor integrated navigation system is based on the known variance of measurement noise, but the variance of measurement noise will change with internal and external interference. Therefore, this paper firstly extends the variational Bayesian approximation based adaptive Kalman filter (VB-AKF) from a single integrated navigation system to a multi-sensor integrated navigation system. Then, two kinds of centralized fusion algorithms of multi-sensor integrated navigation system are proposed, namely, the VB-AKF based augmented centralized fusion algorithm and the VB-AKF based sequential centralized fusion algorithm, to solve the problem of information fusion of multi-sensor integrated navigation with unknown measurement noise variance. Finally, the SINS / GNSS / CNS / ADS multi-sensor integrated navigation system is used to validate the above algorithm. The experimental results show that the two algorithms proposed in this paper have the same filtering accuracy and are close to the ideal centralized Kalman fusion algorithm ( ICKF) when the variance of measurement noise is known. In the whole simulation period, compared with traditional centralized Karl filter (TCKF) and federal Kalman filter (FT-FKF) with fault tolerance function, the proposed algorithm can improve position accuracy by 32% and 90%, and speed accuracy by 38% and 71%, respectively

Abstract:As an important component of the new energy vehicle battery, the quality of the collector tray is related to the performance of the battery and has an important impact on the life safety of the vehicle occupants. In practical industrial applications, real-time detection of battery collector tray defects with limited computational resources is a challenging task. In order to reduce the model size and computational effort, and to reduce the application cost, this paper proposes a lightweight new energy vehicle battery collector tray defect detection model (SGCNet). First, ShuffleNet V2 is used as the backbone feature extraction network, and group convolution and channel shuffle techniques are adopted to reduce the computational complexity and the number of parameters while extracting effective features. Secondly, a lightweight feature fusion network GC-FPN is designed with lightweight GhostNet and CARAFE upsampling operators to fully retain the semantic information of the feature map while reducing parameter redundancy and ensuring detection accuracy, thus reducing the computational cost. The experimental results show that the SGCNet model achieves 90. 6% detection accuracy, the model size is 3. 2 M, the GFLOPs are only 3. 6, and the FPS reaches 178. 6. Compared with the current advanced lightweight network models, it has higher detection accuracy and lower computational effort. Finally, the SGCNet model is deployed on the embedded platform NVIDIA Jetson Nano for real-time detection, with a detection time of 0. 07 s per image, meeting the requirements for accuracy and real-time performance for battery collector defect detection tasks in real industry.

Abstract:Aiming at the problem that the target size of the UAV aerial image is too small and contains less feature information, which leads to the unsatisfactory detection effect of the existing detection algorithm on small objects, a UAV aerial photography based on feature aggregation and multi-collaborative feature interaction is proposed. First of all, in view of the insufficient feature extraction of the backbone network, Swin Transformer is selected as the RetinaNet backbone network to enhance the global information extraction ability of the algorithm. Secondly, in order to improve the detection ability of remote targets, a small target feature aggregation network is proposed, which can fully integrate the details of small targets in shallow feature maps. Finally, in order to further improve the detection performance of multi-scale targets, a new multiple collaborative feature interaction module is proposed to make the low-level feature information flow to the high-level. Experimental results on VisDrone2019-DET, a public UAV aerial photo data set, show that compared with the original RetinaNet baseline network detection precision increased by 7. 6%, the proposed algorithm has better detection effect for small targets.

Abstract:In the anomaly detection of industrial surfaces, due to the unknown and irregular nature of the abnormalities, it is difficult and costly to manually label abnormal samples, and the supervised deep learning algorithms have limitations in the task of anomaly detection on the surface of industrial products. To address the above problems, an unsupervised surface anomaly detection algorithm based on contrastive learning generative adversarial network (CLGAN) is proposed. Firstly, the CLGAN model based on unsupervised learning algorithm is established. Secondly, contrastive learning is used to strengthen the positive and negative sample constraints of the potential feature space, maximizing the mutual information between the corresponding patches of the input and output images, enhancing the differentiation of positive and negative sample feature vectors, and further improving the ability of the model to reconstruct abnormal sample images. Then, in the detection stage, the trained model is used to obtain the anomaly-free reconstruction image of the industrial product to be tested, and the residual image between the sample to be measured and its corresponding reconstructed image is calculated. Finally, combined with the double threshold segmentation method and mathematical morphology processing, the rapid detection and accurate location of abnormal areas on the surface of industrial products are realized. Experimental results on the public dataset MVTec AD demonstrate that the proposed algorithm has a better recognition effect and stronger generalization ability compared with other unsupervised deep learning model algorithms.

Abstract:The energy management system of hydrogen fuel cell vehicles plays an important role in the performance of fuel cell lifespan, durability, and economy. On the basis of improving the classic fuzzy control strategy, the problem of frequent load changes in the output power of fuel cells in the control strategy is solved. The optimization design is carried out using temporal rules, and a control strategy based on temporal period sampling is proposed based on the operating characteristics, efficiency points, and road conditions of fuel cells. By controlling the output power of vehicles equipped with fuel cells and power batteries, complete the optimization of energy management strategies. An analysis was conducted on the hydrogen consumption, fuel cell output power status, and power cell charge state changes of a fuel cell bus in actual operation under constant speed conditions and typical urban bus cycle conditions in China, using different control strategies. It was found that the designed control strategy can effectively reduce the number of fuel cell load changes, improve the service life of the fuel cell, and compare the hydrogen consumption with the power following strategy, the switch control strategy has obvious advantages. Under the initial state of charge of the power battery of 60%, the hydrogen consumption of the latter two strategies increased by 3. 98% and 27. 88%, respectively.

Abstract:In order to achieve contactless and automated online measurement of internal thread parameters, this paper proposes a machine vision measurement system for internal thread pitch based on the spherical catadioptric panoramic imaging principle, using through-hole nuts and blind-hole nuts as inspection objects. Firstly, the system acquires the image obtained by the spherical catadioptric system and segments the complete internal thread area. Secondly, contrast limited adaptive histogram equalization algorithm is used to enhance the image contrast, and a combination of median filtering and bilateral filtering is used to protect the thread boundary information. Then, a Zernike moment edge detection algorithm is used to determine the sub-pixel edges of each thread. Finally, the internal thread pitch dimensions are calculated based on the theory of spherical catadioptric imaging. The pitch measurement values were compared with those of a comprehensive thread measuring machine for metrology. It shows that the system has an average measurement error of 0. 018 5 mm that meets the requirements for accuracy of internal thread pitch measurement in industrial production. The experiments proved that the system is highly effective in detecting and can be used for online visual inspection of internal threads. This study provides a reference solution for cylindrical internal wall dimension measurement and defect detection.

Abstract:In industrial construction, bolt components are key connectors commonly used to join large machinery and components such as steel structures, bridges, highways, buildings, and oil pipelines. The quality of their installation directly affects the stability and reliability of the entire equipment or structure. However, the installation of bolts often takes place in narrow and complex environments, making manual inspection difficult, inefficient, and prone to misjudgment and omissions. In this study, we conducted bolt component recognition research based on the Faster R-CNN framework, aiming to address the challenges in bolt detection, we propose a detection algorithm based on multi-scale and multi-directional bolts. Firstly, we augment the collected images to enhance the diversity of the dataset. Secondly, we enhance the sensitivity of the model to feature information by modifying the backbone network, and utilize a multiscale fusion module to improve the detection of small targets. In the stage of generating bounding boxes, we introduce an adaptive rotation region proposal network to obtain optimal bounding boxes. Finally, we address the issue of discontinuous boundaries in multidirectional detection by employing the Gaussian Wasserstein distance and focal loss as the loss functions instead of the traditional Smooth L1 loss. Experimental results for bolt component recognition demonstrate that the improved Faster R-CNN model achieves a mAP (mean average precision ) value of 87. 4%, which is a 7. 6% improvement over the original Faster R-CNN model. Through ablation experiments, it is observed that the improved ResNet50 network achieves a 0. 2% increase in AP (average precision) compared to the original ResNet50 network. Comparisons with other rotation detection models on the same dataset reveal that the proposed model has higher AP values and better robustness. The model presented in this study effectively addresses the challenges posed by the shooting angles and complex environments in bolt component recognition tasks, mitigating issues caused by image scale and discontinuous rotation boundaries.

Abstract:Aiming at the current problems of cumbersome operation, low efficiency and large variation in repeated measurements of highvoltage cable insulation layer quality inspection, a new type of cable insulation layer inspection device is designed, and a high-voltage cable insulation layer image segmentation method based on improved U-Net is proposed. Firstly, the backbone feature extraction network is replaced with the VGG16 network, the weights trained by VGG16 in the Pascal VOC2012 dataset are used as the pre-training weights in combination with the transfer learning, the adaptive feature weighting mechanism is incorporated in the jump connections by using the channel attention module, as well as the grouped convolution is added in the up-sampling process, which improves the semantic segmentation accuracy. Next, the insulating layer image segmentation is performed using the trained optimal weights, the contour region features are extracted and binarised, and the contour region is filled using the connected region algorithm. Finally, the complete insulation layer segmentation image is generated by fusing the original image and the segmented region. The experimental results show that the mean intersection-over-union and mean pixel accuracy reach 99. 56% and 99. 81%, which is a significant improvement over the original network effect, and verifies the effectiveness of the method on the segmentation of the insulation layer of high-voltage cables.