Heat tracing cable detection method and system based on image processing

Abstract

The invention discloses a heat tracing cable detection method and system based on image processing, and relates to the technical field of image processing. Comprising the following steps: carrying out image acquisition by adopting various types of sensors, and integrating obtained image data; according to the integrated data, a dynamic background model is constructed through an adaptive image enhancement technology, and a heat tracing cable and a complex background are separated through a background modeling method; performing feature extraction by using a convolutional neural network, automatically identifying key features of the heat tracing cable, and adapting to a new detection task by using a pre-trained network model through a transfer learning technology; the visible light camera, the thermal infrared imager and the laser scanner are integrated, information of the cable is captured from different angles and dimensions and integrated into a comprehensive cable state view, then multi-scale image data fusion technologies including image preprocessing, feature point matching algorithm alignment and image pyramid technologies are utilized, the detection efficiency is improved, and the detection precision is improved. And the accuracy of a detection result is also remarkably improved.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a heating cable detection method and system based on image processing. Background Art

[0002] In the power system, heating cables are important equipment used to prevent pipelines, containers and other equipment from freezing and to maintain process temperatures. Traditional heating cable detection methods mainly rely on manual visual inspection or simple temperature measurement. These methods have problems such as low detection efficiency, poor accuracy, and inability to monitor in real time. With the development of image processing technology, image-based detection methods are gradually being used in the detection of heating cables due to their non-contact, high efficiency and high precision. However, existing technologies still have limitations in dealing with complex backgrounds, noise interference and accurately identifying defects. Summary of the invention

[0003] The purpose of the present invention is to provide a heating cable detection method and system based on image processing.

[0004] The purpose of the present invention can be achieved through the following technical solutions:

[0005] The present application provides a heating cable detection method based on image processing, comprising the following steps:

[0006] Using multiple types of sensors to collect images and integrate the acquired image data; the multiple types of sensors include visible light cameras, infrared thermal imagers and laser scanners, the infrared thermal imagers are used to capture temperature distribution, and the laser scanners are used to obtain the three-dimensional shape of the heating cable;

[0007] Based on the integrated data, a dynamic background model is constructed through adaptive image enhancement technology, and the heating cable and complex background are separated through background modeling method;

[0008] Use convolutional neural networks for feature extraction to automatically identify key features of heating cables, and use transfer learning technology to use pre-trained network models to adapt to new detection tasks;

[0009] The method of using a convolutional neural network to extract features and automatically identify key features of the heating cable specifically includes:

[0010] Feature extraction is performed through a convolution operation, which is expressed as:

[0011] V = conv2(W, X, "valid") + b, where V is the output after the convolution operation, b is the bias term, conv2 represents a two-dimensional convolution operation, and "valid" represents a valid convolution without edge padding;

[0012] The convolutional neural network learns complex features through the activation function, and the output of the activation function is expressed as: Where Y is the output of the activation function, represents the activation function;

[0013] The spatial size of the feature map is reduced by the pooling operation, which is expressed as: Where Z is the output of the pooling layer, Y is the output of the convolutional layer, r is the size of the pooling window, i, j represent the position of the pooling window, and u and v are used to specify the row and column indices of the elements within the pooling window;

[0014] After the convolution, activation and pooling layers, the classification and output are performed through the fully connected layer.

[0015] Furthermore, the obtained image data is integrated through multi-scale image data fusion technology, specifically including:

[0016] Preprocess images collected by different sensors to correct lighting differences, remove noise and enhance image quality;

[0017] When integrating images from visible light, infrared, and laser scanning, feature point matching algorithms are used to identify and align key features in the image, followed by feature alignment.

[0018] Use image pyramid technology to fuse features of different scales;

[0019] The aligned and fused image data are integrated to create a composite data set containing multiple information sources, and a weighted average is used to combine temperature, shape and image information to generate a comprehensive picture of the heating cable condition.

[0020] Furthermore, the image pyramid technology is used to fuse features of different scales, including: using a weighted average method to fuse and generate a fused feature set, the fused feature set is expressed as: Where N represents the total number of sensors, indicating how many different sensors are involved in feature extraction; F i represents the feature set extracted by the i-th sensor. The features extracted by each sensor may contain different information; α i The weight coefficient of the i-th sensor feature is used to indicate the importance of the sensor feature in the fusion process, satisfying

[0021] Furthermore, a dynamic background model is constructed using a generative adversarial network, including:

[0022] Construct a generative adversarial network, including a generator and a discriminator, where the generator is used to receive random noise z and the background information I of the current framet , generate a background image B similar to the current image t ; The generator output form is: B t =G(z,I t );

[0023] The discriminator is used to receive the background information I of the current frame of the real image t and the generated background image B t , to determine whether the input image is from a real dataset or an image generated by the generator, the discriminator output is: D(I t ) and D(B t ); where D(I t ) is the judgment of the real image, and D(B t ) is the judgment of the generated image;

[0024] The discriminator is trained through the adversarial loss function, and then the generated background image is matched with the real background image at the pixel level through the loss function of the generator combined with the pixel-level reconstruction loss;

[0025] The adversarial loss function is expressed as:

[0026] LD=-E[log D(I t )]-E[log(1-D(G(z,I t )))]; where -E[logD(I t )] is the prediction loss of the discriminator for the real image, -E[log(1-D(G(z,I t )))] is the prediction loss for the generated image;

[0027] The loss function of the generator is expressed as:

[0028] LG = -E[log D(G(z,I t ))]+λ·L r , where L G is the total loss of the generator, E[logD(G(z,I t ))] is the expected logarithmic loss of the image generated by the generator that is judged by the discriminator as a real image, λ is a hyperparameter used to balance the weight between adversarial loss and reconstruction loss, and L r is the pixel-level reconstruction loss, which is used to measure the difference between the generated background image and the real background image;

[0029] In each round of training, the parameters of the discriminator and the generator are optimized alternately. By fixing the parameters of the generator and updating the parameters of the discriminator to minimize the discriminator loss, a realistic background image is generated.

[0030] The generator generates a background image based on the current input image and random noise, and extracts the foreground by calculating the difference between the current image and the generated background. The generated background image is expressed as:

[0031] B t =G(z,I t ), the extraction prospect is expressed as: F t =I t -B t , where F t The foreground image of the heating cable is extracted.

[0032] Furthermore, after automatically identifying the key features of the heating cable, the method further includes: using an image segmentation algorithm to accurately segment the cable.

[0033] Furthermore, the use of an image segmentation algorithm to accurately segment the cable specifically includes:

[0034] The original cable image is converted into the HSV color space, the image is analyzed using the Mask R-CNN deep learning model, candidate regions containing cables are generated through a region proposal network, and pixel-level segmentation is performed; the region proposal network generates candidate regions and corresponding feature maps;

[0035] For each candidate region proposed by the region proposal network, Mask R-CNN predicts whether the region contains a cable and performs pixel-level classification to determine whether each pixel belongs to a cable. Through the calculation of the convolutional layer and the fully connected layer, an accurate mask of each candidate region is generated for pixel-level segmentation of the cable.

[0036] After the segmentation is completed, the generated mask is used to correct the segmentation error and fill small holes.

[0037] Furthermore, it also includes: based on the identified features, using the YOLO algorithm to detect and classify cable defects, classifying the detected defects into different types, and providing real-time feedback on the defect level and severity;

[0038] Furthermore, the cable defects are detected and classified using the YOLO algorithm, including:

[0039] The YOLO algorithm is used to build a deep learning model to detect cable defects. The trained YOLO model is used to perform real-time detection of cable images in actual scenes. The bounding box and category probability are predicted by a single network unit for rapid detection of cable defects. The specific expression is as follows:

[0040] Output m，n，j =(c, d, w, h, p object , p class1 , pclass2 , ..., p classN ), where (c, d) represents the coordinates of the center of the bounding box relative to the upper left corner of the grid cell, which is normalized relative to the size of the grid cell, (w, h) represents the width and height of the bounding box, which is normalized relative to the size of the entire image, and p object Indicates the confidence that the bounding box contains the object, and the confidence that the target object does exist in the bounding box, p class1 , p class2 , ..., p classN is the conditional probability of each category, which represents the conditional probability of each category within the bounding box.

[0041] Furthermore, it also includes: according to the type of defects detected and their severity, the detection results are correlated with historical data for analysis, and maintenance suggestions and preventive measures are automatically generated.

[0042] A heating cable detection system based on image processing is used to implement a heating cable detection method based on image processing, specifically comprising:

[0043] The data acquisition and fusion module is used to collect image data from different sensors and integrate the acquired image data through multi-scale image data fusion technology;

[0044] The dynamic background modeling module uses a generative adversarial network to build a dynamic background model, including a generator and a discriminator. The generator is responsible for generating a background image similar to the current frame, while the discriminator distinguishes between real images and generated images. The parameters are adaptively adjusted through adversarial training to adapt to environmental changes in real time.

[0045] The defect detection and classification module automatically identifies the key features of the heating cable based on the convolutional neural network and the YOLO algorithm, and performs defect detection and classification. It outputs the prediction results through multi-layer convolution, activation, pooling and full connection operations to quickly detect cable defects.

[0046] The maintenance suggestion and decision support module associates the inspection results with historical data and automatically generates maintenance suggestions and preventive measures based on the type and severity of the defects detected.

[0047] The beneficial effects of the present invention are:

[0048] The present invention captures detailed information of the cable from different angles and dimensions, including temperature distribution and three-dimensional shape, by integrating multiple sensors such as visible light cameras, infrared thermal imagers and laser scanners. The integration of these data provides a comprehensive view of the cable status. Multi-scale image data fusion technology, including image preprocessing, feature point matching algorithm alignment and image pyramid technology, is then used to effectively integrate feature sets from multiple sensors, improve the expressiveness of features, and thus obtain better performance and robustness in tasks such as image recognition and target detection. The present invention solves the limitations of traditional heating cable detection methods, especially the problems of low detection efficiency, poor accuracy and inability to monitor in real time, which not only improves the detection efficiency, but also significantly improves the accuracy of the detection results.

[0049] A dynamic background model is constructed using a generative adversarial network (GAN), which effectively separates the heating cable from the complex background, improves the usability of the image, and can adaptively adjust parameters under different environmental conditions to adapt to environmental changes in real time. In addition, the cable defect detection and classification is carried out through the YOLO algorithm, which realizes the rapid detection of cable defects and provides real-time feedback on the defect level and severity, helping staff to prioritize serious problems. Finally, according to the detected defect type and its severity, it automatically performs correlation analysis with historical data, generates maintenance recommendations and preventive measures, and helps operators to formulate effective maintenance plans. This not only realizes real-time monitoring of the heating cable, but also provides intelligent decision-making support for the safe and stable operation of the power system. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] For better understanding and implementation, the technical solution of the present application is described in detail below with reference to the accompanying drawings.

[0051] Figure 1 A schematic flow chart of a heating cable detection method based on image processing provided in Example 1 of the present application;

[0052] Figure 2 A schematic flow chart of integrating acquired image data in a heating cable detection method based on image processing provided in Example 1 of the present application;

[0053] Figure 3 A schematic diagram of a process of constructing a dynamic background model using a generative adversarial network in a heating cable detection method based on image processing provided in Example 1 of the present application;

[0054] Figure 4 A schematic diagram of the feature extraction process of a heating cable detection method based on image processing using a convolutional neural network provided in Example 1 of the present application. DETAILED DESCRIPTION

[0055] In order to further explain the technical means and effects taken by the present invention to achieve the predetermined invention purpose, exemplary embodiments will be described in detail here, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are only examples of methods and systems consistent with some aspects of the present application as detailed in the attached claims.

[0056] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0057] The specific implementation methods, features and effects of the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments.

[0058] Example 1

[0059] See also Figure 1-Figure 4 This embodiment provides a heating cable detection method and system based on image processing, which comprehensively monitors and detects the heating cable with higher accuracy and efficiency, adapts to complex environmental conditions, provides real-time feedback and intelligent decision support, and thus significantly improves the maintenance and management level of the heating cable.

[0060] The present invention provides a heating cable detection method based on image processing, comprising the following steps:

[0061] S1. Use multiple types of sensors to collect images and integrate the obtained image data; the multiple types of sensors include visible light cameras, infrared thermal imagers and laser scanners. The infrared thermal imager is used to capture temperature distribution, and the laser scanner is used to obtain the three-dimensional shape of the heating cable to enhance the comprehensive monitoring of the cable status;

[0062] Furthermore, the obtained image data is integrated through multi-scale image data fusion technology, specifically including:

[0063] S11, preprocessing images collected by different sensors to correct illumination differences, remove noise and enhance image quality;

[0064] S12, when integrating images from visible light, infrared and laser scanning, identify and align key features in the image by using a feature point matching algorithm, and then perform feature alignment;

[0065] S13, using image pyramid technology to fuse features of different scales;

[0066] Furthermore, the image pyramid technology is used to fuse features of different scales, including: using a weighted average method to fuse and generate a fused feature set, the fused feature set is expressed as: Where N represents the total number of sensors, indicating how many different sensors are involved in feature extraction; F i represents the feature set extracted by the i-th sensor. The features extracted by each sensor may contain different information; α i The weight coefficient of the i-th sensor feature is used to indicate the importance of the sensor feature in the fusion process, satisfying

[0067] Specifically, the use of image pyramid technology combined with the weighted average method for feature fusion can effectively integrate feature sets extracted from multiple sensors at different scales. This method assigns a weight coefficient to each sensor feature. These weight coefficients reflect the relative importance of each sensor feature in the fusion process, and the sum of all weight coefficients must be equal to 1 to ensure the normalization of the weights. In this way, the generated fusion feature set can combine the advantages of each sensor and improve the expressiveness of the features, thereby achieving better performance and robustness in tasks such as image recognition and target detection.

[0068] S14. Integrate the aligned and fused image data to create a composite data set containing multiple information sources, and then use weighted averaging to combine the temperature, shape and image information to generate a comprehensive heating cable status image.

[0069] Specifically, by using multiple types of sensors for image acquisition and integration, including visible light cameras, infrared thermal imagers and laser scanners, the status of the heating cable can be fully monitored. Through multi-scale image data fusion technology, the acquired images are first pre-processed to correct lighting differences, remove noise and enhance image quality, and then the key features in the image are aligned using a feature point matching algorithm. Then, the features of different scales are fused using image pyramid technology and weighted averaging methods to generate a fused feature set; this method can combine the advantages of each sensor and improve the expression ability of the features. Finally, by integrating the processed image data, a composite data set containing temperature, shape and image information is created to generate a comprehensive heating cable status image, thereby achieving better performance and robustness in tasks such as image recognition and target detection.

[0070] S2. Based on the integrated data, a dynamic background model is constructed through adaptive image enhancement technology. The heating cable and the complex background are effectively separated through the background modeling method to improve the usability of the image. In this way, the parameters can be adaptively adjusted under different environmental conditions to adapt to environmental changes in real time;

[0071] Furthermore, a dynamic background model is constructed using a generative adversarial network, including:

[0072] S21, construct a generative adversarial network, including a generator and a discriminator, where the generator is used to receive random noise z and background information I of the current frame t , generate a background image B similar to the current image t ; The generator output form is: B t =G(z,I t );

[0073] The discriminator is used to receive the background information I of the current frame of the real image t and the generated background image B t , to determine whether the input image is from a real dataset or an image generated by the generator, the discriminator output is: D(I t ) and D(B t ); where D(I t ) is the judgment of the real image, and D(B t ) is the judgment of the generated image;

[0074] S22, training the discriminator through the adversarial loss function so that it can accurately distinguish between the real image and the generated image, and then matching the generated background image with the real background image at the pixel level through the loss function of the generator combined with the pixel-level reconstruction loss;

[0075] The adversarial loss function is expressed as:

[0076] L D = -E[logD(I t )]-E[log(1-D(G(z,I t )))]; where -E|log D(I t )] is the prediction loss of the discriminator for the real image, -E[log(1-D(G(z,I t )))] is the prediction loss for the generated image;

[0077] The loss function of the generator is expressed as:

[0078] L G = -E[log D(G(z,I t ))]+λ·L r, where LG is the total loss of the generator, E[log D(G(z,I t ))] is the expected logarithmic loss of the image generated by the generator that is judged by the discriminator as a real image, λ is a hyperparameter used to balance the weight between adversarial loss and reconstruction loss, and L r is the pixel-level reconstruction loss, which is used to measure the difference between the generated background image and the real background image;

[0079] S23. In each round of training, the parameters of the discriminator and the generator are optimized alternately. First, the parameters of the generator are fixed, and the parameters of the discriminator are updated to minimize the discriminator loss. This adversarial training allows the generator to continuously improve and gradually generate more realistic background images.

[0080] The generator generates a background image based on the current input image and random noise, and extracts the foreground by calculating the difference between the current image and the generated background. The generated background image is expressed as:

[0081] B t =G(z,I t ), the extraction prospect is expressed as: F t =I t -B t , where F t The foreground image of the heating cable is extracted.

[0082] Specifically, through the application of adaptive image enhancement technology and generative adversarial networks, we can effectively build a dynamic background model and separate the heating cable from the complex background, thereby improving the usability of the image. In this process, the generator receives random noise and the background information of the current frame to generate a background image similar to the current image, while the discriminator determines whether the input image is real or generated. The discriminator is trained through the adversarial loss function so that it can accurately distinguish between real images and generated images. At the same time, combined with the pixel-level reconstruction loss, it ensures that the generated background image matches the real background image at the pixel level. In each round of training, the parameters of the discriminator and the generator are optimized alternately. This adversarial training enables the generator to continuously improve and gradually generate more realistic background images. Finally, the foreground, that is, the foreground image of the heating cable, is extracted by calculating the difference between the current image and the generated background, which realizes the adaptive adjustment of parameters under different environmental conditions to adapt to environmental changes in real time, thereby improving the accuracy and robustness of heating cable detection.

[0083] S3. Use convolutional neural network (CNN) for feature extraction to automatically identify the key features of heating cables. Through transfer learning technology, use pre-trained network models to adapt to new detection tasks and significantly improve recognition accuracy.

[0084] Furthermore, a convolutional neural network (CNN) is used for feature extraction to automatically identify the key features of the heating cable, including:

[0085] S31, performing feature extraction through convolution operation, wherein the convolution operation is expressed as:

[0086] V = conv2(W, X, "valid") + b, where V is the output after the convolution operation, b is the bias term, conv2 represents a two-dimensional convolution operation, and "valid" represents a valid convolution without edge padding;

[0087] S32, enabling the convolutional neural network to learn complex features through an activation function, wherein the output of the activation function is expressed as: Where Y is the output of the activation function, represents the activation function;

[0088] S33, reducing the spatial size of the feature map through pooling operation, reducing the number of parameters and the amount of calculation, and extracting important features at the same time, the pooling operation is expressed as: Where Z is the output of the pooling layer, Y is the output of the convolutional layer, r is the size of the pooling window, i, j represent the position of the pooling window, and u and v are used to specify the row and column indices of the elements within the pooling window;

[0089] S34, after convolution, activation and pooling layers, classification and output are performed through the fully connected layer.

[0090] Specifically, in the process of using convolutional neural networks (CNN) to extract features and automatically identify key features of heating cables, the output is the prediction results after multi-layer convolution, activation, pooling and full connection operations. These results are usually high-level representations of key features in the heating cable image and can be used for subsequent classification, positioning or detection tasks, thereby realizing automatic identification and analysis of heating cables.

[0091] Furthermore, after automatically identifying the key features of the heating cable, the method also includes: using an image segmentation algorithm to accurately segment the cable to ensure the accuracy of defect detection.

[0092] Furthermore, the use of an image segmentation algorithm to accurately segment the cable specifically includes:

[0093] The original cable image is converted into the HSV color space, the image is analyzed using the MaskR-CNN deep learning model, candidate regions containing cables are generated through a region proposal network, and pixel-level segmentation is performed; the region proposal network generates candidate regions and corresponding feature maps;

[0094] For each candidate region proposed by the region proposal network, Mask R-CNN further predicts whether the region contains a cable and performs pixel-level classification to determine whether each pixel belongs to a cable. Through the calculation of the convolutional layer and the fully connected layer, an accurate mask for each candidate region is generated to achieve pixel-level segmentation of the cable.

[0095] After the segmentation is completed, the generated mask is used to correct the segmentation error and fill small holes.

[0096] Specifically, by converting the original cable image into HSV color space and applying the Mask R-CNN deep learning model, we are able to achieve accurate segmentation of the cable. This process first uses the region proposal network (RPN) to identify and generate candidate regions containing cables and their feature maps in the image. Then Mask R-CNN classifies these regions at the pixel level and predicts whether each pixel belongs to a cable. Through the calculation of the convolutional layer and the fully connected layer, the model generates an accurate mask for each candidate region, thereby achieving pixel-level segmentation of the cable. Finally, the generated mask is post-processed to correct the segmentation error and fill small holes to ensure the accuracy and completeness of the segmentation results, providing strong technical support for cable detection and analysis.

[0097] S4. Based on the identified features, the cable defects are detected and classified through the YOLO algorithm, and the detected defects are classified into different types (such as insulation damage, temperature abnormality, poor contact, etc.), and the defect level and severity are fed back in real time to help staff prioritize serious problems;

[0098] Furthermore, the cable defects are detected and classified using the YOLO algorithm, including:

[0099] The YOLO algorithm is used to build a deep learning model to detect cable defects. The trained YOLO model is used to perform real-time detection of cable images in actual scenes. The bounding box and category probability are predicted by a single network unit to achieve rapid detection of cable defects. The specific expression is as follows:

[0100] Output m，n，j =(c, d, w, h, p object , p class1 , p class2 , ..., p classN ), where (c, d) represents the coordinates of the center of the bounding box relative to the upper left corner of the grid cell, which is normalized relative to the size of the grid cell, (w, h) represents the width and height of the bounding box, which is normalized relative to the size of the entire image, and p object Indicates the confidence that the bounding box contains the object, and the confidence that the target object does exist in the bounding box, p class1 , pclass2 , ..., p classN is the conditional probability of each category, which represents the conditional probability of each category within the bounding box.

[0101] Specifically, the YOLO model divides the image into grid cells, each of which is responsible for predicting the object whose center point falls within the grid. The deep learning model built using the YOLO algorithm can perform real-time and accurate defect detection on cable images in actual scenarios. The model predicts the center coordinates, width, height, and confidence of the object contained in each bounding box through a single network unit, and calculates the conditional probability of belonging to different categories for each target object in the bounding box. This detection method is not only fast, but also can provide accurate defect location and classification information, allowing staff to quickly identify and prioritize serious defects in cables, thereby improving the efficiency and safety of cable maintenance.

[0102] S5. According to the type of defects detected and their severity, the detection results are correlated with historical data for analysis, and maintenance suggestions and preventive measures are automatically generated to help operators develop effective maintenance plans.

[0103] The heating cable detection method based on image processing provided in this embodiment realizes comprehensive monitoring of the state of the heating cable by integrating multiple sensor data and applying multi-scale image data fusion technology. First, the system uses multiple types of sensors such as visible light cameras, infrared thermal imagers and laser scanners for image acquisition, and fuses features of different scales through image pyramid technology and weighted average method to generate a fusion feature set. Then, a dynamic background model is constructed using a generative adversarial network (GAN) to effectively separate the heating cable from the complex background and improve the usability of the image. Then, feature extraction is performed through a convolutional neural network (CNN) to automatically identify the key features of the heating cable, and with the help of transfer learning technology, a pre-trained network model is used to adapt to new detection tasks, significantly improving the recognition accuracy. Further, an image segmentation algorithm such as Mask R-CNN is used to accurately segment the cable to ensure the accuracy of defect detection. Finally, based on the identified features, a deep learning model such as YOLO is used to detect and classify the cable defects, and the detection results are correlated with historical data for analysis, and maintenance suggestions and preventive measures are automatically generated to help operators formulate effective maintenance plans. It realizes comprehensive, efficient and precise monitoring of heating cables, significantly improves the accuracy and efficiency of detection, adapts to complex environmental conditions, provides real-time feedback and intelligent decision-making support, and thus greatly improves the maintenance and management level of heating cables.

[0104] Example 2

[0105] A heating cable detection system based on image processing is used to implement a heating cable detection method based on image processing, specifically comprising:

[0106] The data acquisition and fusion module is used to collect image data from different sensors and integrate the acquired image data through multi-scale image data fusion technology;

[0107] The dynamic background modeling module uses a generative adversarial network (GAN) to build a dynamic background model, including a generator and a discriminator. The generator is responsible for generating background images similar to the current frame, while the discriminator distinguishes between real images and generated images. Through adversarial training, the parameters are adaptively adjusted to adapt to environmental changes in real time.

[0108] The defect detection and classification module, based on the convolutional neural network (CNN) and YOLO algorithm, automatically identifies the key features of the heating cable, and performs defect detection and classification. It outputs the prediction results through multi-layer convolution, activation, pooling and full connection operations, quickly detects cable defects, and provides accurate defect location and classification information;

[0109] The maintenance suggestion and decision support module associates the inspection results with historical data and automatically generates maintenance suggestions and preventive measures based on the type and severity of the defects detected.

[0110] This embodiment solves the problems of low efficiency, poor accuracy and inability to monitor in real time in traditional heating cable detection methods through data acquisition and fusion modules, dynamic background modeling modules, defect detection and classification modules and maintenance suggestions and decision support modules. Through the collaborative work of visible light cameras, infrared thermal imagers and laser scanners, the system can fully monitor the cable status and accurately identify and classify cable defects through image processing technology. In addition, the system can adapt to different environmental conditions, provide real-time feedback on defect information, and provide intelligent decision support, thereby significantly improving the maintenance and management level of heating cables and ensuring the safe and stable operation of the power system.

[0111] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technical personnel in this field can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any brief modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A heating cable detection method based on image processing, characterized in that: The steps include: Using multiple types of sensors to collect images and integrate the acquired image data; the multiple types of sensors include visible light cameras, infrared thermal imagers and laser scanners, the infrared thermal imagers are used to capture temperature distribution, and the laser scanners are used to obtain the three-dimensional shape of the heating cable; Based on the integrated data, a dynamic background model is constructed through adaptive image enhancement technology, and the heating cable and complex background are separated through background modeling method; Use convolutional neural networks for feature extraction to automatically identify key features of heating cables, and use transfer learning technology to use pre-trained network models to adapt to new detection tasks; The method of using a convolutional neural network to extract features and automatically identify key features of the heating cable specifically includes: Feature extraction is performed through a convolution operation, which is expressed as: V = conv2(W, X, "valid") + b, where V is the output after the convolution operation, b is the bias term, conv2 represents a two-dimensional convolution operation, and "valid" represents a valid convolution without edge padding; The convolutional neural network learns complex features through the activation function, and the output of the activation function is expressed as: Where Y is the output of the activation function, represents the activation function; The spatial size of the feature map is reduced by the pooling operation, which is expressed as: Where Z is the output of the pooling layer, Y is the output of the convolutional layer, r is the size of the pooling window, i, j represent the position of the pooling window, and u and v are used to specify the row and column indices of the elements within the pooling window; After the convolution, activation and pooling layers, the classification and output are performed through the fully connected layer.

2. A heating cable detection method based on image processing according to claim 1, characterized in that: The acquired image data is integrated through multi-scale image data fusion technology, including: Preprocess images collected by different sensors to correct lighting differences, remove noise and enhance image quality; When integrating images from visible light, infrared, and laser scanning, feature point matching algorithms are used to identify and align key features in the image, followed by feature alignment. Use image pyramid technology to fuse features of different scales; The aligned and fused image data are integrated to create a composite data set containing multiple information sources, and a weighted average is used to combine temperature, shape and image information to generate a comprehensive picture of the heating cable condition.

3. The method for detecting a heating cable based on image processing according to claim 2 is characterized in that: Using the image pyramid technology, features of different scales are fused, including: using a weighted average method to fuse and generate a fused feature set, the fused feature set is expressed as: Where N represents the total number of sensors, indicating how many different sensors are involved in feature extraction; F i represents the feature set extracted by the i-th sensor. The features extracted by each sensor may contain different information; α i The weight coefficient of the i-th sensor feature is used to indicate the importance of the sensor feature in the fusion process, satisfying 4. The method for detecting a heating cable based on image processing according to claim 1, characterized in that: Use generative adversarial networks to build a dynamic background model, including: Construct a generative adversarial network, including a generator and a discriminator, where the generator is used to receive random noise z and the background information I of the current frame t , generate a background image B similar to the current image t ; The generator output form is: B t =G(z,I t ); The discriminator is used to receive the background information I of the current frame of the real image t and the generated background image B t , to determine whether the input image is from a real dataset or an image generated by the generator, the discriminator output is: D(I t ) and D(B t ); where D(I t ) is the judgment of the real image, and D(B t ) is the judgment of the generated image; The discriminator is trained through the adversarial loss function, and then the generated background image is matched with the real background image at the pixel level through the loss function of the generator combined with the pixel-level reconstruction loss; The adversarial loss function is expressed as: L D = -E[log D(I t )]-E[log(1-D(G(z,I t )))]; where -E[log D(I t )] is the prediction loss of the discriminator for the real image, -E[log(1-D(G(z,I t )))] is the prediction loss for the generated image; The loss function of the generator is expressed as: L G = -E[log D(G(z,I t ))]+λ·L r , where L G is the total loss of the generator, E[log D(G(z,I t ))] is the expected logarithmic loss of the image generated by the generator that is judged by the discriminator as a real image, λ is a hyperparameter used to balance the weight between adversarial loss and reconstruction loss, and L r is the pixel-level reconstruction loss, which is used to measure the difference between the generated background image and the real background image; In each round of training, the parameters of the discriminator and the generator are optimized alternately. By fixing the parameters of the generator and updating the parameters of the discriminator to minimize the discriminator loss, a realistic background image is generated. The generator generates a background image based on the current input image and random noise, and extracts the foreground by calculating the difference between the current image and the generated background. The generated background image is expressed as: B t =G(z,I t ), the extraction prospect is expressed as: F t =I t -B t , where F t The foreground image of the heating cable is extracted.

5. The method for detecting a heating cable based on image processing according to claim 1, characterized in that: After automatically identifying the key features of the heating cable, it also includes: using image segmentation algorithms to accurately segment the cable.

6. A heating cable detection method based on image processing according to claim 5, characterized in that: The use of an image segmentation algorithm to accurately segment the cable specifically includes: The original cable image is converted into the HSV color space, the image is analyzed using the Mask R-CNN deep learning model, candidate regions containing cables are generated through a region proposal network, and pixel-level segmentation is performed; the region proposal network generates candidate regions and corresponding feature maps; For each candidate region proposed by the region proposal network, Mask R-CNN predicts whether the region contains a cable and performs pixel-level classification to determine whether each pixel belongs to a cable. Through the calculation of the convolutional layer and the fully connected layer, an accurate mask of each candidate region is generated for pixel-level segmentation of the cable. After the segmentation is completed, the generated mask is used to correct the segmentation error and fill small holes.

7. The method for detecting heating cables based on image processing according to claim 1, characterized in that: Also includes: Based on the identified features, the cable defects are detected and classified through the YOLO algorithm, the detected defects are divided into different types, and the defect level and severity are fed back in real time.

8. The method for detecting heating cables based on image processing according to claim 7 is characterized in that: The cable defects are detected and classified by the YOLO algorithm, including: The YOLO algorithm is used to build a deep learning model to detect cable defects. The trained YOLO model is used to perform real-time detection of cable images in actual scenes. The bounding box and category probability are predicted by a single network unit for rapid detection of cable defects. The specific expression is as follows: Output m，n，j =(c, d, w, h, p object , p class1 , p class2 , ..., p classN ), where (c, d) represents the coordinates of the center of the bounding box relative to the upper left corner of the grid cell, which is normalized relative to the size of the grid cell, (w, h) represents the width and height of the bounding box, which is normalized relative to the size of the entire image, and p object Indicates the confidence that the bounding box contains the object, and the confidence that the target object does exist in the bounding box, p class1 , p class2 , ..., p classN is the conditional probability of each category, which represents the conditional probability of each category within the bounding box.

9. The method for detecting heating cables based on image processing according to claim 1, characterized in that: Also includes: Based on the type of defects detected and their severity, the inspection results are correlated with historical data and maintenance recommendations and preventive measures are automatically generated.

10. A heating cable detection system based on image processing, used to implement a heating cable detection method based on image processing as claimed in any one of claims 1 to 9, characterized in that: include: The data acquisition and fusion module is used to collect image data from different sensors and integrate the acquired image data through multi-scale image data fusion technology; The dynamic background modeling module uses a generative adversarial network to build a dynamic background model, including a generator and a discriminator. The generator is responsible for generating a background image similar to the current frame, while the discriminator distinguishes between real images and generated images. The parameters are adaptively adjusted through adversarial training to adapt to environmental changes in real time. The defect detection and classification module automatically identifies the key features of the heating cable based on the convolutional neural network and the YOLO algorithm, and performs defect detection and classification. It outputs the prediction results through multi-layer convolution, activation, pooling and full connection operations to quickly detect cable defects. The maintenance suggestion and decision support module associates the inspection results with historical data and automatically generates maintenance suggestions and preventive measures based on the type and severity of the defects detected.

Images