Power transmission line external damage prevention monitoring method based on deep learning target detection and tracking

Abstract

The invention discloses a power transmission line external damage prevention monitoring method based on deep learning target detection and tracking, relates to the technical field of power transmission line external damage prevention monitoring, and aims to solve the problems of poor image data acquisition quality of a scene around a power transmission line and inaccurate foreign matter recognition and tracking. According to the method, the processing speed is optimized through the YOLOv6 model, real-time monitoring of the power transmission line can be achieved, the external damage risk can be quickly responded, the YOLOv6 model can adapt to complex scenes and illumination changes, the recognition capacity in different environments is improved, the whole process automation from data preprocessing, model reasoning to result recognition is achieved, and the recognition efficiency is improved. The DeepSORT model can process complex conditions such as disappearance and re-occurrence of a target, supports real-time processing and can detect and track the target in a monitoring video stream in real time in power transmission line external damage prevention monitoring due to environmental factors and diversity of potential targets.

Description

Technical Field

[0001] The present invention relates to the technical field of power transmission line anti-breakage monitoring technology, and specifically to a power transmission line anti-breakage monitoring method based on deep learning target detection and tracking. Background Art

[0002] Transmission line anti-external damage monitoring refers to the use of a series of technical means and methods to monitor and protect transmission lines to prevent line damage or failures caused by external factors.

[0003] The patent application with announcement number CN105744232B discloses a method for preventing external damage to power transmission lines based on behavior analysis technology. It can adapt to the needs of preventing external damage to power facilities in various scenarios by adjusting system parameters. In indoor scenarios, personnel intrusion detection and behavior recognition analysis are realized. In outdoor scenarios, real-time recognition and analysis of the movement status and behavior type (intrusion, wandering, stopping, raising arms, etc.) of large construction vehicles such as cranes and excavators are realized. When an external damage event of power facilities occurs in the detection area, it can accurately identify and quickly respond to the alarm, and push alarm text messages, pictures and alarm videos through the wireless network to avoid serious damage to power facilities. Although the above patent solves the problem of preventing external damage, the following problems still exist in actual operation: 1. The acquired image data is not further processed for image quality, resulting in poor analysis efficiency when the image is analyzed later.

[0004] 2. The acquired image data is not used to perform target detection and foreign object tracking using a better model, resulting in an inability to more accurately capture the scene around the transmission line.

[0005] 3. There is no more careful data analysis of the scenes around the transmission lines, and no targeted warnings based on abnormal events, which leads to the inability to better resolve abnormal events. Summary of the invention

[0006] The purpose of the present invention is to provide a transmission line anti-external damage monitoring method based on deep learning target detection and tracking. The YOLOv6 model is optimized in processing speed, so that real-time monitoring of transmission lines can be achieved and external damage risks can be quickly responded to. The YOLOv6 model can adapt to complex scenes and lighting changes, improve the recognition ability in different environments, and realize the automation of the whole process from data preprocessing, model reasoning to result recognition. The DeepSORT model can handle complex situations such as the disappearance and reappearance of targets. In the transmission line anti-external damage monitoring, due to the diversity of environmental factors and potential targets, the DeepSORT model supports real-time processing, can detect and track targets in real time in the monitoring video stream, and can solve the problems in the prior art.

[0007] To achieve the above object, the present invention provides the following technical solutions: The transmission line anti-external damage monitoring method based on deep learning target detection and tracking includes the following steps: S1: Transmission line data collection: Use detection equipment to collect image data and video data of the transmission line, and mark them as pending data after collection; S2: Collected data processing: Process the collected data to be processed, and obtain the target collected data after the data processing is completed; S3: Target data detection: Use the YOLOv6 model to perform real-time image recognition on the target collected data, and mark the recognized data as target foreign body data; S4: Detection data tracking: Use the DeepSORT model to track the target foreign body data, and mark the tracking data as standard tracking data; S5: Tracking data evaluation: Perform abnormal analysis on the standard tracking data, determine abnormal events based on the abnormal analysis results, and mark the abnormal events as transmission line monitoring data; S6: Evaluation data warning and storage: Make warning judgments on abnormal events in the transmission line monitoring data, issue an alarm based on the judgment results, generate monitoring reports for the alarm data and the transmission line monitoring data respectively, and transmit them to the display terminal.

[0008] Preferably, the detection device in S1 collects image data and video data of the power transmission line, including: The detection equipment is a drone, which uses the camera on board to collect image and video data of the scene around the transmission line; Before the drone collects data about the scene around the power transmission line, the drone is first tested for performance; The performance test of UAV includes appearance inspection, electrical system inspection, camera inspection, sensor inspection, flight performance inspection and communication inspection; After the performance test of the UAV is qualified, data collection is carried out, and the collected data is marked as pending collected data.

[0009] Preferably, the data to be processed in S2 is processed, including: Compressing the video data in the collected data to be processed, and decoding the compressed video into a plurality of image frames; According to a fixed interval, a number of image frames are extracted, wherein the fixed interval time is set by oneself; The images extracted from the video data and the image data collected by the drone are uniformly processed for image quality; Image quality processing includes image denoising, white balance adjustment and sharpening; First, the extracted image and image data are filtered using a median filter model, and the filter parameters are adjusted according to the filtering effect; The image after the filter parameter adjustment is subjected to white balance adjustment. The white balance adjustment is to detect the brightness value of the filtered image, adjust the brightness value according to the brightness value detection result, adjust the gray value of the image after the brightness value adjustment, and perform color rendering after the gray value adjustment is completed; The foreground and background areas of the color-rendered image are identified, and the contrast between the foreground and background areas is calculated. If the contrast is less than the set contrast threshold, the foreground area is enhanced, and the edge features of the critical position between the foreground area and the background area are extracted. The width of the critical position is identified based on the edge features. If the width of the critical position exceeds the set width threshold, the width of the critical position is symmetrically reduced on both sides to be equal to the width threshold. After the critical position width of the image extracted from the video data is adjusted, a plurality of image frames are subjected to video integration, and after the integration is completed, video data with image quality processing is obtained; The video data and image data after image quality processing are statistically marked as target acquisition data.

[0010] Preferably, the brightness value adjustment includes: Extracting a brightness value corresponding to the image data, and comparing the brightness value with a preset brightness threshold; When the brightness value is lower than the brightness threshold, the grayscale values ​​of all pixels included in the image data are extracted; Obtaining a brightness adjustment coefficient using the grayscale values ​​of all pixels included in the image data; The brightness adjustment coefficient is obtained by the following formula:

[0011] Where S represents the brightness adjustment coefficient; n represents the number of pixels contained in the image data; H i Indicates the gray value corresponding to the i-th pixel; H z Indicates the central gray value corresponding to the image data; H xz represents the grayscale median value corresponding to n pixels; L represents the brightness value corresponding to the image data; L y Indicates the brightness threshold; The brightness adjustment coefficient is used to adjust the brightness of the image data whose brightness value is lower than the brightness threshold, and the brightness value of the image data after the brightness adjustment is obtained by the following formula:

[0012] Among them, L t Indicates the brightness value of the image data after brightness adjustment.

[0013] Preferably, performing image enhancement processing on the foreground area includes: Extracting the grayscale value before grayscale adjustment and the grayscale value after grayscale adjustment corresponding to each pixel point included in the foreground area; Obtaining the grayscale adjustment amplitude ratio corresponding to each pixel point by using the grayscale value before grayscale adjustment and the grayscale value after grayscale adjustment corresponding to each pixel point; Extracting the brightness value of the foreground area before and after the brightness adjustment; Obtaining a brightness adjustment amplitude ratio corresponding to the foreground area by using the brightness value of the foreground area before brightness adjustment and the brightness value of the foreground area after brightness adjustment; The contrast adjustment coefficient of the foreground area is obtained by combining the grayscale adjustment amplitude ratio corresponding to each pixel point with the brightness adjustment amplitude ratio corresponding to the foreground area; The contrast adjustment coefficient of the foreground area is obtained by the following formula:

[0014] Where Y represents the contrast adjustment coefficient of the foreground area; m represents the number of pixels contained in the foreground area; P hi P represents the grayscale adjustment ratio of the i-th pixel contained in the foreground area; l Indicates the brightness adjustment ratio of the foreground area; P b Indicates the brightness adjustment ratio of the background area; D x Indicates the contrast value corresponding to the foreground area; D b Indicates the contrast value corresponding to the background area; D y Indicates the set contrast threshold; The contrast of the foreground area is adjusted using the contrast adjustment coefficient, wherein the adjusted contrast value corresponding to the foreground area is obtained by the following formula:

[0015] Among them, D t represents the adjusted contrast value corresponding to the foreground area; Y represents the contrast adjustment coefficient of the foreground area.

[0016] Preferably, the target acquisition data in S3 is subjected to real-time image recognition using a YOLOv6 model, including: Retrieve the YOLOv6 model from the database and train the YOLOv6 model. After the model training is completed, use the deep learning framework to load the weights. After the weights are loaded, a standard YOLOv6 model is obtained. Preprocessing the target acquisition data, including image size adjustment, image pixel normalization and image data enhancement; The target acquisition data after data preprocessing is input into the standard YOLOv6 model for forward reasoning. Forward reasoning is to call the reasoning function of the standard YOLOv6 model. The target acquisition data after data preprocessing is input into the reasoning function to obtain the model output. The model output includes the target category, the bounding box coordinates of each target and its confidence score. The model output results are processed for recognition. The recognition process is to filter out the expected targets according to the set confidence threshold, remove the detection results with low confidence, and apply the NMS strategy to the overlapping prediction boxes to eliminate duplicate detections. After recognition processing, the data with recognition anomalies in the target collected data is obtained and marked as target foreign body data.

[0017] Preferably, in S4, the target foreign body data is tracked using the DeepSORT model, including: The DeepSORT model is retrieved from the database and trained. After the model training is completed, the model is initialized. The initialization process includes configuring the feature extractor, the overlap threshold between the detection box and the tracking box, and the confidence threshold for screening the target. After the model initialization process, a standard DeepSORT model is obtained. The target foreign body data is subjected to feature extraction, and the feature extraction includes the bounding box, category label and appearance feature vector of the target foreign body data, wherein the bounding box is the position of the target in the image, the category label is the type information of the detected target, and the appearance feature vector is the feature extraction of the target bounding box using a convolutional neural network and the generation of the appearance feature vector.

[0018] Preferably, the method of using the DeepSORT model to track the target foreign body data in S4 further includes: The bounding box and appearance feature vector of the target foreign body data are input into the standard DeepSORT model. The standard DeepSORT model uses a Kalman filter to predict the motion state of the target, where the motion state includes position, speed and direction; The standard DeepSORT model then matches the currently detected target with the already existing targets based on the similarity metric; The trajectory is updated according to the matching results, wherein the trajectory is updated using the update step of the Carl filter for the successfully matched target; for the unmatched detection results, a new trajectory is initialized; for the unmatched trajectory, the matching count is increased, and the trajectory is deleted after exceeding the preset number; After the trajectory update is completed, the target tracking result in each frame of the target foreign object data is input, and the target tracking result includes the position, speed and direction of the foreign object; Finally, the target tracking results are labeled as standard tracking data.

[0019] Preferably, performing an abnormal analysis on the standard tracking data in S5 and determining an abnormal event according to the abnormal analysis result include: First, define abnormal events into abnormal indicators, including speed abnormality, trajectory deviation, stagnation detection and target loss; Standard tracking data is analyzed for anomalies using real-time monitoring, model analysis, and cluster analysis; Real-time monitoring is to monitor each target through defined abnormal indicators in each frame of standard tracking data; model analysis is to use deep learning models to judge standard tracking data; cluster analysis is to use clustering algorithms to analyze the movement pattern of the target; The three abnormal analysis results are fused to obtain abnormal events in the standard tracking data, and the abnormal events are marked as transmission line monitoring data.

[0020] Preferably, in S6, an early warning judgment is made for abnormal events in the transmission line monitoring data, an alarm is issued according to the judgment result, and monitoring reports are generated for the alarm data and the transmission line monitoring data respectively and transmitted to the display terminal, including: Conduct abnormal event analysis on the transmission line monitoring data, including analyzing the specific type, frequency and duration of abnormal events; The analyzed abnormal events are divided into severity according to the degree of abnormality, and the severity is divided into slight abnormality, moderate abnormality and severe abnormality; The warning judgment is made according to the severity of the classification. When the threshold of the severity of the classification is triggered to the warning threshold, the warning alarm is generated. Different severities correspond to different degrees of warning. Visualize the analyzed abnormal events and generated warning alerts using visualization tools; After visualization processing, the data is transmitted to the display terminal for report display.

[0021] Compared with the prior art, the present invention has the following beneficial effects: 1. The power transmission line anti-external damage monitoring method based on deep learning target detection and tracking provided by the present invention converts video data into image frames through video compression and decoding, which provides convenience for subsequent processing and reduces the burden of data storage and transmission. Image extraction is performed according to a set fixed interval, which effectively reduces the amount of data processed while retaining key information, improving processing efficiency, and extracting edge features between foreground and background, which helps to more accurately identify the position and shape of the target.

[0022] 2. The present invention provides a transmission line anti-external damage monitoring method based on deep learning target detection and tracking. The YOLOv6 model is optimized in processing speed, and can realize real-time monitoring of transmission lines and quickly respond to external damage risks. The YOLOv6 model can adapt to complex scenes and lighting changes, improve recognition capabilities in different environments, and realize the automation of the entire process from data preprocessing, model reasoning to result recognition. The DeepSORT model can handle complex situations such as the disappearance and reappearance of targets. In the transmission line anti-external damage monitoring, due to the diversity of environmental factors and potential targets, the DeepSORT model supports real-time processing and can detect and track targets in real time in the monitoring video stream.

[0023] 3. The power transmission line anti-external damage monitoring method based on deep learning target detection and tracking provided by the present invention combines real-time monitoring, model analysis and cluster analysis, and can conduct in-depth analysis of standard tracking data from different angles, improve the accuracy and reliability of anomaly detection, and automate the process from abnormal event analysis to early warning judgment, alarm generation and monitoring report generation, which reduces manual intervention and errors and improves the intelligent level of operation and maintenance management. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a schematic diagram of the steps of monitoring the power transmission line against external damage according to the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] In order to solve the problem in the prior art that after the scene around the power transmission line is collected, the collected image data is not further processed for image quality, resulting in poor analysis efficiency in the later image analysis, please refer to Figure 1 , this embodiment provides the following technical solutions: The transmission line anti-external damage monitoring method based on deep learning target detection and tracking includes the following steps: S1: Transmission line data collection: Use detection equipment to collect image data and video data of the transmission line, and mark them as pending data after collection; The entire process from drone performance testing to data collection, labeling, processing and analysis can be highly automated, helping to reduce labor costs; S2: Collected data processing: Process the collected data to be processed, and obtain the target collected data after the data processing is completed; Among them, through video compression and decoding, video data is converted into image frames, which facilitates subsequent processing and reduces the burden of data storage and transmission; S3: Target data detection: Use the YOLOv6 model to perform real-time image recognition on the target collected data, and mark the recognized data as target foreign body data; Among them, the YOLOv6 model can adapt to complex scenes and lighting changes, improve recognition capabilities in different environments, and realize the automation of the entire process from data preprocessing, model reasoning to result recognition, reducing manual intervention and improving monitoring efficiency; S4: Detection data tracking: Use the DeepSORT model to track the target foreign body data, and mark the tracking data as standard tracking data; Among them, DeepSORT combines Kalman filtering and the Hungarian algorithm to accurately track the target and continue tracking it when the target is occluded or disappears and then reappears; S5: Tracking data evaluation: Perform abnormal analysis on the standard tracking data, determine abnormal events based on the abnormal analysis results, and mark the abnormal events as transmission line monitoring data; Among them, the combination of real-time monitoring, model analysis and cluster analysis can conduct in-depth analysis of standard tracking data from different angles and improve the accuracy and reliability of anomaly detection; S6: Evaluation data warning and storage: Make warning judgments on abnormal events in the transmission line monitoring data, issue an alarm based on the judgment results, generate monitoring reports for the alarm data and the transmission line monitoring data, and transmit them to the display terminal; Among them, the automated process from abnormal event analysis to early warning judgment, alarm generation, and monitoring report generation reduces manual intervention and errors and improves the intelligence level of operation and maintenance management.

[0027] The detection equipment in S1 collects image data and video data of the transmission line, including: The detection equipment is a drone, which uses the camera on board to collect image and video data of the scene around the transmission line; Before the drone collects data about the scene around the power transmission line, the drone is first tested for performance; The performance test of UAV includes appearance inspection, electrical system inspection, camera inspection, sensor inspection, flight performance inspection and communication inspection; After the performance test of the UAV is qualified, data collection is carried out, and the collected data is marked as pending collected data.

[0028] Specifically, drones can quickly fly over complex terrain and conduct all-round, no-dead-angle inspections of power transmission lines, greatly improving inspection efficiency. Compared with traditional manual inspections, drone inspections are not only faster, but can also cover more areas that are difficult to reach manually. The cameras on drones have high-definition monitoring capabilities and can accurately capture scene details around power transmission lines, providing high-quality image and video data for deep learning target detection and tracking. This helps to more accurately identify and analyze potential sources of external damage risks. Before data collection, detailed performance testing of drones, including appearance inspection, electrical system testing, camera testing, sensor testing, flight performance testing, and communication testing, can ensure the stability and reliability of drones during inspections, thereby ensuring the accuracy and integrity of data collection. From drone performance testing to data collection, labeling, processing, and analysis, the entire process can be highly automated, which helps to reduce labor costs and improve the overall efficiency of the monitoring system.

[0029] Data processing is performed on the collected data to be processed in S2, including: Compressing the video data in the collected data to be processed, and decoding the compressed video into a plurality of image frames; According to a fixed interval, a number of image frames are extracted, wherein the fixed interval time is set by oneself; The images extracted from the video data and the image data collected by the drone are uniformly processed for image quality; Image quality processing includes image denoising, white balance adjustment and sharpening; First, the extracted image and image data are filtered using a median filter model, and the filter parameters are adjusted according to the filtering effect; The image after the filter parameter adjustment is subjected to white balance adjustment. The white balance adjustment is to detect the brightness value of the filtered image, adjust the brightness value according to the brightness value detection result, adjust the gray value of the image after the brightness value adjustment, and perform color rendering after the gray value adjustment is completed; The foreground and background areas of the color-rendered image are identified, and the contrast between the foreground and background areas is calculated. If the contrast is less than the set contrast threshold, the foreground area is enhanced, and the edge features of the critical position between the foreground area and the background area are extracted. The width of the critical position is identified based on the edge features. If the width of the critical position exceeds the set width threshold, the width of the critical position is symmetrically reduced on both sides to be equal to the width threshold. After the critical position width of the image extracted from the video data is adjusted, a plurality of image frames are subjected to video integration, and after the integration is completed, video data with image quality processing is obtained; The video data and image data after image quality processing are statistically marked as target acquisition data.

[0030] Specifically, through video compression and decoding, video data is converted into image frames, which provides convenience for subsequent processing and reduces the burden of data storage and transmission. Image extraction is performed according to a set fixed interval, which effectively reduces the amount of data to be processed, while retaining key information and improving processing efficiency. The median filter model is used for filtering, which effectively removes noise in the image and improves the clarity of the image. White balance adjustment is performed according to the brightness value of the image to ensure the accuracy and consistency of the image color and improve the readability of the image. Through sharpening processing, the details of the image are enhanced to make the image clearer. The identification of the foreground and background areas of the image is helpful for subsequent target detection and tracking. By adjusting the contrast of the foreground and background areas and performing image enhancement processing on the foreground area, the accuracy of target detection is improved. At the same time, the edge features between the foreground and background are extracted, which helps to more accurately identify the position and shape of the target. By adjusting the contrast of the foreground and background areas and performing image enhancement processing on the foreground area, the accuracy of target detection is improved. At the same time, extracting edge features between the foreground and background helps to more accurately identify the position and shape of the target, reintegrate the processed image frames into video data, maintain the continuity and integrity of the original video, and perform statistical annotation on the processed video data and image data, providing rich data support for subsequent target detection and tracking.

[0031] Specifically, the brightness value adjustment includes: Extracting a brightness value corresponding to the image data, and comparing the brightness value with a preset brightness threshold; When the brightness value is lower than the brightness threshold, the grayscale values ​​of all pixels included in the image data are extracted; Obtaining a brightness adjustment coefficient using the grayscale values ​​of all pixels included in the image data; The brightness adjustment coefficient is obtained by the following formula:

[0032] Where S represents the brightness adjustment coefficient; n represents the number of pixels contained in the image data; H i Indicates the gray value corresponding to the i-th pixel; H z Indicates the central gray value corresponding to the image data; H xz represents the grayscale median value corresponding to n pixels; L represents the brightness value corresponding to the image data; L y Indicates the brightness threshold; The brightness adjustment coefficient is used to adjust the brightness of the image data whose brightness value is lower than the brightness threshold, and the brightness value of the image data after the brightness adjustment is obtained by the following formula:

[0033] Among them, L t Indicates the brightness value of the image data after brightness adjustment.

[0034] The technical effect of the above technical solution is: by comparing the brightness value of the image data with the preset brightness threshold, the solution can automatically identify the image with insufficient brightness and adjust the brightness thereof in a targeted manner. This adaptive method avoids indiscriminate processing of all images, thereby improving the efficiency and accuracy of processing. When the image brightness is lower than the threshold, the scheme extracts the grayscale values ​​of all pixels in the image and uses these grayscale values ​​to calculate the brightness adjustment coefficient. This method takes into account the brightness information of each pixel in the image and can achieve more precise and accurate brightness adjustment. The calculation formula of the brightness adjustment coefficient S comprehensively considers multiple factors, including the grayscale value of the pixel, the center grayscale value, the grayscale median value, the original brightness value and the brightness threshold. This dynamic calculation method makes the brightness adjustment more flexible and adaptable, and can provide a suitable brightness adjustment scheme according to different image conditions. The brightness value of the image data after brightness adjustment can be accurately calculated by the formula, which makes the brightness effect after adjustment predictable and evaluated in advance. This is very important for application scenarios that require strict control of image brightness. Through the above brightness adjustment method, the visual effect of images with insufficient brightness can be significantly improved, making them clearer and easier to observe. At the same time, since the grayscale information of the pixels is taken into account, the adjusted image will also be improved in terms of maintaining details and contrast. This solution calculates the brightness adjustment coefficient by comprehensively considering multiple factors and uses a precise formula to calculate the adjusted brightness value, which makes the algorithm more robust and stable when facing different images and lighting conditions.

[0035] In summary, the technical effects of the above technical solutions in terms of performance indicators are mainly reflected in adaptive brightness adjustment, fine adjustment based on pixel grayscale values, dynamic calculation of brightness adjustment coefficients, predictable effects after brightness adjustment, improved image quality, and enhanced algorithm robustness, etc. These effects jointly improve the efficiency and accuracy of image processing, and provide strong support for applications in fields such as image processing and computer vision.

[0036] Specifically, the foreground area is subjected to image enhancement processing, including: Extracting the grayscale value before grayscale adjustment and the grayscale value after grayscale adjustment corresponding to each pixel point included in the foreground area; Obtaining the grayscale adjustment amplitude ratio corresponding to each pixel point by using the grayscale value before grayscale adjustment and the grayscale value after grayscale adjustment corresponding to each pixel point; Extracting the brightness value of the foreground area before and after the brightness adjustment; Obtaining a brightness adjustment amplitude ratio corresponding to the foreground area by using the brightness value of the foreground area before brightness adjustment and the brightness value of the foreground area after brightness adjustment; The contrast adjustment coefficient of the foreground area is obtained by combining the grayscale adjustment amplitude ratio corresponding to each pixel point with the brightness adjustment amplitude ratio corresponding to the foreground area; The contrast adjustment coefficient of the foreground area is obtained by the following formula:

[0037] Where Y represents the contrast adjustment coefficient of the foreground area; m represents the number of pixels contained in the foreground area; P hi P represents the grayscale adjustment ratio of the i-th pixel contained in the foreground area; l Indicates the brightness adjustment ratio of the foreground area; P b Indicates the brightness adjustment ratio of the background area; D x Indicates the contrast value corresponding to the foreground area; D b Indicates the contrast value corresponding to the background area; D y Indicates the set contrast threshold; The contrast of the foreground area is adjusted using the contrast adjustment coefficient, wherein the adjusted contrast value corresponding to the foreground area is obtained by the following formula:

[0038] Among them, D t represents the adjusted contrast value corresponding to the foreground area; Y represents the contrast adjustment coefficient of the foreground area.

[0039] The technical effect of the above technical solution is: the solution can achieve fine adjustment of the contrast of the foreground area by calculating the grayscale adjustment amplitude ratio of each pixel and the brightness adjustment amplitude ratio of the foreground area. This fine adjustment helps to maintain the detail information in the image and avoid image distortion caused by excessive enhancement or reduction of contrast. The calculation formula of the contrast adjustment coefficient Y comprehensively considers multiple factors, including the grayscale adjustment amplitude ratio of the pixel, the brightness adjustment amplitude ratio of the foreground and background areas, the contrast value of the foreground area, and the set contrast threshold. This adaptive adjustment method makes the contrast enhancement effect more in line with the actual needs of the image and improves the visual effect of the image. When calculating the contrast adjustment coefficient, the scheme also considers the brightness adjustment amplitude ratio and contrast value of the background area. This helps to maintain the contrast balance between the foreground and the background while enhancing the contrast of the foreground area, avoiding the foreground being too prominent and the background being too dim or blurred. By calculating the adjusted contrast value Dt by the formula, the effect of contrast adjustment can be predicted in advance. This is very important for application scenarios where image contrast needs to be strictly controlled, and helps to ensure that the adjusted image meets specific visual requirements. Contrast is one of the important indicators of image quality. Through fine and adaptive contrast adjustment, the scheme can significantly improve the overall quality of the image, making the image clearer, more vivid and easier to observe. The scheme uses flexible formulas and parameter settings, which can be adjusted and optimized according to different images and application scenarios. This enhances the flexibility and adaptability of the algorithm, enabling it to cope with a variety of complex image enhancement needs.

[0040] In summary, the technical effects of the above technical solutions in terms of performance indicators are mainly reflected in fine contrast adjustment, adaptive contrast enhancement, balanced contrast between foreground and background, predictable contrast adjustment effect, improved overall image quality, and enhanced algorithm flexibility and adaptability, etc. These effects jointly improve the efficiency and accuracy of image enhancement processing, and provide strong support for applications in fields such as image processing and computer vision.

[0041] In order to solve the problem in the prior art that the acquired image data is not used for target detection and foreign object tracking with a better model, which leads to the inability to more accurately obtain the scene around the transmission line, please refer to Figure 1 , this embodiment provides the following technical solutions: Use the YOLOv6 model to perform real-time image recognition on the target data collected in S3, including: Retrieve the YOLOv6 model from the database and train the YOLOv6 model. After the model training is completed, use the deep learning framework to load the weights. After the weights are loaded, a standard YOLOv6 model is obtained. Preprocessing the target acquisition data, including image size adjustment, image pixel normalization and image data enhancement; The target acquisition data after data preprocessing is input into the standard YOLOv6 model for forward reasoning. Forward reasoning is to call the reasoning function of the standard YOLOv6 model. The target acquisition data after data preprocessing is input into the reasoning function to obtain the model output. The model output includes the target category, the bounding box coordinates of each target and its confidence score. The model output results are processed for recognition. The recognition process is to filter out the expected targets according to the set confidence threshold, remove the detection results with low confidence, and apply the NMS strategy to the overlapping prediction boxes to eliminate duplicate detections. After recognition processing, the data with recognition anomalies in the target collected data is obtained and marked as target foreign body data.

[0042] Specifically, the YOLOv6 model has been optimized in terms of processing speed, and can realize real-time monitoring of transmission lines and quickly respond to external damage risks. After weight loading through the deep learning framework, the YOLOv6 model can quickly complete forward reasoning and output the target category, bounding box coordinates and its confidence score, providing the possibility for timely preventive measures. The YOLOv6 model has achieved excellent performance on multiple data sets, and can accurately identify foreign objects in transmission lines and reduce false alarm and missed alarm rates. Through data enhancement technology, the YOLOv6 model can adapt to complex scenes and lighting changes, improve recognition capabilities in different environments, and realize the automation of the entire process from data preprocessing, model reasoning to result recognition, reducing manual intervention and improving monitoring efficiency. According to the set confidence threshold, the system can automatically screen out the expected targets and apply the NMS strategy to the overlapping prediction boxes to eliminate duplicate detections and realize intelligent early warning. The YOLOv6 model provides models of various sizes, ranging from lightweight versions to standard versions. You can choose the appropriate model for deployment according to actual needs.

[0043] In S4, the target foreign body data is tracked using the DeepSORT model, including: The DeepSORT model is retrieved from the database and trained. After the model training is completed, the model is initialized. The initialization process includes configuring the feature extractor, the overlap threshold between the detection box and the tracking box, and the confidence threshold for screening the target. After the model initialization process, a standard DeepSORT model is obtained. The target foreign body data is subjected to feature extraction, and the feature extraction includes the bounding box, category label and appearance feature vector of the target foreign body data, wherein the bounding box is the position of the target in the image, the category label is the type information of the detected target, and the appearance feature vector is the feature extraction of the target bounding box using a convolutional neural network and the generation of the appearance feature vector.

[0044] The bounding box and appearance feature vector of the target foreign body data are input into the standard DeepSORT model. The standard DeepSORT model uses a Kalman filter to predict the motion state of the target, where the motion state includes position, speed and direction; The standard DeepSORT model then matches the currently detected target with the already existing targets based on the similarity metric; The trajectory is updated according to the matching results, wherein the trajectory is updated using the update step of the Carl filter for the successfully matched target; for the unmatched detection results, a new trajectory is initialized; for the unmatched trajectory, the matching count is increased, and the trajectory is deleted after exceeding the preset number; After the trajectory update is completed, the target tracking result in each frame of the target foreign object data is input, and the target tracking result includes the position, speed and direction of the foreign object; Finally, the target tracking results are labeled as standard tracking data.

[0045] Specifically, the DeepSORT model supports multi-target tracking, which means that it can track multiple foreign targets at the same time, such as multiple objects that may affect the safety of transmission lines. This is particularly important in the monitoring of transmission line damage prevention, because there may be multiple potential sources of danger at the same time. The DeepSORT model combines Kalman filtering and the Hungarian algorithm to accurately track the target and continue tracking when the target is occluded or disappears and then reappears. It uses powerful appearance feature descriptors to accurately distinguish the similarities between different targets, which helps to improve tracking accuracy. The DeepSORT model can handle complex situations such as the disappearance and reappearance of targets, which is crucial for long-term monitoring and tracking tasks. In the monitoring of power line damage prevention, due to environmental factors (such as weather changes, lighting conditions, etc.) and the diversity of potential targets, the tracking algorithm needs to be highly robust. The DeepSORT model supports real-time processing and can detect and track targets in real time in the monitoring video stream. It is of great significance for timely warning and emergency measures, and helps to reduce power transmission line failures caused by external force damage. Before using the DeepSORT model, the model can be trained and initialized according to specific scenarios and needs, including configuring feature extractors, setting the overlap threshold between detection boxes and tracking boxes, and the confidence threshold for screening targets. This adaptability enables the DeepSORT model to better adapt to different monitoring environments and target types.

[0046] In order to solve the problem that the existing technology does not conduct more careful data analysis on the scenes around the transmission line, and does not provide targeted warnings based on abnormal events, which leads to the inability to better solve the abnormal events, please refer to Figure 1 , this embodiment provides the following technical solutions: Perform anomaly analysis on the standard tracking data in S5 and judge abnormal events based on the anomaly analysis results, including: First, define abnormal events into abnormal indicators, including speed abnormality, trajectory deviation, stagnation detection and target loss; Standard tracking data is analyzed for anomalies using real-time monitoring, model analysis, and cluster analysis; Real-time monitoring is to monitor each target through defined abnormal indicators in each frame of standard tracking data; model analysis is to use deep learning models to judge standard tracking data; cluster analysis is to use clustering algorithms to analyze the movement pattern of the target; The three abnormal analysis results are fused to obtain abnormal events in the standard tracking data, and the abnormal events are marked as transmission line monitoring data.

[0047] Specifically, firstly, abnormal indicators such as speed anomaly, trajectory deviation, stagnation detection and target loss are defined. These indicators can fully cover various abnormal situations that may be encountered in transmission line monitoring, thereby ensuring the comprehensiveness of abnormal detection. The combination of real-time monitoring, model analysis and cluster analysis can conduct in-depth analysis of standard tracking data from different angles to improve the accuracy and reliability of anomaly detection. Real-time monitoring can timely detect anomalies in the current frame, model analysis can use the powerful judgment ability of deep learning models to evaluate the overall data, and cluster analysis can reveal the movement pattern of the target and further assist in abnormal judgment. The application of deep learning models enables this scheme to more accurately identify abnormal events in transmission line monitoring. Through a large amount of data training, the deep learning model can learn the characteristics of transmission line monitoring data, thereby realizing accurate judgment of abnormal events. Real-time monitoring can monitor abnormal indicators for each target in each frame of standard tracking data, so as to timely detect abnormal situations and respond. This real-time monitoring method can greatly improve the timeliness of transmission line anti-external damage monitoring and reduce losses caused by delayed response. Cluster analysis can use clustering algorithms to analyze the movement pattern of the target, thereby revealing potential abnormal behaviors. This analysis method can provide powerful auxiliary information for abnormal judgment, help the system to judge abnormal situations more accurately, and integrate the three abnormal analysis results. It can make full use of the advantages of various analysis methods and improve the accuracy of abnormal judgment.

[0048] In S6, an abnormal event in the transmission line monitoring data is pre-warned and judged, an alarm is issued according to the judgment result, and monitoring reports are generated for the alarm data and the transmission line monitoring data respectively and transmitted to the display terminal, including: Conduct abnormal event analysis on the transmission line monitoring data, including analyzing the specific type, frequency and duration of abnormal events; The analyzed abnormal events are divided into severity according to the degree of abnormality, and the severity is divided into slight abnormality, moderate abnormality and severe abnormality; The warning judgment is made according to the severity of the classification. When the threshold of the severity of the classification is triggered to the warning threshold, the warning alarm is generated. Different severities correspond to different degrees of warning. Visualize the analyzed abnormal events and generated warning alerts using visualization tools; After visualization processing, the data is transmitted to the display terminal for report display.

[0049] Specifically, through deep learning algorithms, the solution can intelligently classify the analyzed abnormal events into severity categories according to the degree of abnormality, such as minor abnormalities, moderate abnormalities, and severe abnormalities. This helps operation and maintenance personnel quickly identify and handle abnormal events of different severity levels, and can analyze abnormal events in the transmission line monitoring data in real time, including the specific type, frequency, and duration of abnormal events, so as to achieve instant grasp of the status of the transmission line. When the severity of the classified abnormal events triggers the warning threshold, the system will generate a warning alarm and issue alarms of different degrees according to the severity level. This can ensure that operation and maintenance personnel receive important information in a timely manner and take corresponding treatment measures. The analyzed abnormal events can be visualized using visualization tools. The generated early warning alarms are visualized, such as displayed on the display terminal in the form of charts and images, which helps operation and maintenance personnel to intuitively understand the abnormal conditions of the transmission lines. Visual monitoring reports and data displays enable operation and maintenance personnel to understand the line status more quickly, so as to formulate maintenance plans and response strategies more accurately, and reduce decision-making errors caused by incomplete information or inaccurate understanding. The automated process from abnormal event analysis to early warning judgment, alarm generation, and monitoring report generation reduces manual intervention and errors, and improves the intelligence level of operation and maintenance management. Through automated monitoring and intelligent early warning, the solution can reduce the workload of operation and maintenance personnel, reduce operation and maintenance costs, and improve operation and maintenance efficiency and quality.

[0050] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0051] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

Claims

1. A power transmission line anti-external damage monitoring method based on deep learning target detection and tracking, characterized in that: The steps include: S1: Transmission line data collection: Use detection equipment to collect image data and video data of the transmission line, and mark them as pending data after collection; S2: Collected data processing: Process the collected data to be processed, and obtain the target collected data after the data processing is completed; S3: Target data detection: Use the YOLOv6 model to perform real-time image recognition on the target collected data, and mark the recognized data as target foreign body data; S4: Detection data tracking: Use the DeepSORT model to track the target foreign body data, and mark the tracking data as standard tracking data; S5: Tracking data evaluation: Perform abnormal analysis on the standard tracking data, determine abnormal events based on the abnormal analysis results, and mark the abnormal events as transmission line monitoring data; S6: Evaluation data warning and storage: Make warning judgments on abnormal events in the transmission line monitoring data, issue an alarm based on the judgment results, generate monitoring reports for the alarm data and the transmission line monitoring data, and transmit them to the display terminal; Data processing is performed on the collected data to be processed in S2, including: Compressing the video data in the collected data to be processed, and decoding the compressed video into a plurality of image frames; According to a fixed interval, a number of image frames are extracted, wherein the fixed interval time is set by oneself; The images extracted from the video data and the image data collected by the drone are uniformly processed for image quality; Image quality processing includes image denoising, white balance adjustment and sharpening; First, the extracted image and image data are filtered using a median filter model, and the filter parameters are adjusted according to the filtering effect; The image after the filter parameter adjustment is subjected to white balance adjustment. The white balance adjustment is to detect the brightness value of the filtered image, adjust the brightness value according to the brightness value detection result, adjust the gray value of the image after the brightness value adjustment, and perform color rendering after the gray value adjustment is completed; The foreground and background areas of the color-rendered image are identified, and the contrast between the foreground and background areas is calculated. If the contrast is less than the set contrast threshold, the foreground area is enhanced, and the edge features of the critical position between the foreground area and the background area are extracted. The width of the critical position is identified based on the edge features. If the width of the critical position exceeds the set width threshold, the width of the critical position is symmetrically reduced on both sides to be equal to the width threshold. After the critical position width of the image extracted from the video data is adjusted, a plurality of image frames are subjected to video integration, and after the integration is completed, video data with image quality processing is obtained; The video data and image data after image quality processing are statistically marked as target acquisition data; The brightness value adjustment includes: Extracting a brightness value corresponding to the image data, and comparing the brightness value with a preset brightness threshold; When the brightness value is lower than the brightness threshold, the grayscale values ​​of all pixels included in the image data are extracted; Obtaining a brightness adjustment coefficient using the grayscale values ​​of all pixels included in the image data; The brightness adjustment coefficient is obtained by the following formula: ； Where S represents the brightness adjustment coefficient; n represents the number of pixels contained in the image data; H i Indicates the gray value corresponding to the i-th pixel; H z Indicates the central gray value corresponding to the image data; H xz represents the grayscale median value corresponding to n pixels; L represents the brightness value corresponding to the image data; L y Indicates the brightness threshold; The brightness adjustment coefficient is used to adjust the brightness of the image data whose brightness value is lower than the brightness threshold, and the brightness value of the image data after the brightness adjustment is obtained by the following formula: ； Among them, L t Indicates the brightness value of the image data after brightness adjustment.

2. The power transmission line anti-external damage monitoring method based on deep learning target detection and tracking according to claim 1 is characterized in that: The detection equipment in S1 collects image data and video data of the transmission line, including: The detection equipment is a drone, which uses the camera on board to collect image and video data of the scene around the transmission line; Before the drone collects data about the scene around the power transmission line, the drone is first tested for performance; The performance test of UAV includes appearance inspection, electrical system inspection, camera inspection, sensor inspection, flight performance inspection and communication inspection; After the performance test of the UAV is qualified, data collection is carried out, and the collected data is marked as pending collected data.

3. The power transmission line anti-external damage monitoring method based on deep learning target detection and tracking according to claim 2 is characterized in that: Perform image enhancement on the foreground area, including: Extracting the grayscale value before grayscale adjustment and the grayscale value after grayscale adjustment corresponding to each pixel point included in the foreground area; Obtaining the grayscale adjustment amplitude ratio corresponding to each pixel point by using the grayscale value before grayscale adjustment and the grayscale value after grayscale adjustment corresponding to each pixel point; Extracting the brightness value of the foreground area before and after the brightness adjustment; Obtaining a brightness adjustment amplitude ratio corresponding to the foreground area by using the brightness value of the foreground area before brightness adjustment and the brightness value of the foreground area after brightness adjustment; The contrast adjustment coefficient of the foreground area is obtained by combining the grayscale adjustment amplitude ratio corresponding to each pixel point with the brightness adjustment amplitude ratio corresponding to the foreground area; The contrast adjustment coefficient of the foreground area is obtained by the following formula: ； Where Y represents the contrast adjustment coefficient of the foreground area; m represents the number of pixels contained in the foreground area; P hi P represents the grayscale adjustment ratio of the i-th pixel contained in the foreground area; l Indicates the brightness adjustment ratio of the foreground area; P b Indicates the brightness adjustment ratio of the background area; D x Indicates the contrast value corresponding to the foreground area; D b Indicates the contrast value corresponding to the background area; D y Indicates the set contrast threshold; The contrast of the foreground area is adjusted using the contrast adjustment coefficient, wherein the adjusted contrast value corresponding to the foreground area is obtained by the following formula: ； Among them, D t represents the adjusted contrast value corresponding to the foreground area; Y represents the contrast adjustment coefficient of the foreground area.

4. The method for monitoring power transmission lines against external damage based on deep learning target detection and tracking according to claim 3 is characterized in that: The target data collected in S3 is used for real-time image recognition using the YOLOv6 model, including: Retrieve the YOLOv6 model from the database and train the YOLOv6 model. After the model training is completed, use the deep learning framework to load the weights. After the weights are loaded, a standard YOLOv6 model is obtained. Preprocessing the target acquisition data, including image size adjustment, image pixel normalization and image data enhancement; The target acquisition data after data preprocessing is input into the standard YOLOv6 model for forward reasoning. Forward reasoning is to call the reasoning function of the standard YOLOv6 model. The target acquisition data after data preprocessing is input into the reasoning function to obtain the model output. The model output includes the target category, the bounding box coordinates of each target and its confidence score. The model output results are processed for recognition. The recognition process is to screen out the expected targets according to the set confidence threshold, remove the detection results with low confidence, and apply the NMS strategy to the overlapping prediction boxes to eliminate duplicate detections. After recognition processing, the data with recognition anomalies in the target collected data is obtained and marked as target foreign body data.

5. The method for monitoring power transmission lines against external damage based on deep learning target detection and tracking according to claim 4 is characterized in that: In S4, the target foreign body data is tracked using the DeepSORT model, including: The DeepSORT model is retrieved from the database and trained. After the model training is completed, the model is initialized. The initialization process includes configuring the feature extractor, the overlap threshold between the detection box and the tracking box, and the confidence threshold for screening the target. After the model initialization process, the standard DeepSORT model is obtained. The target foreign body data is subjected to feature extraction, and the feature extraction includes the bounding box, category label and appearance feature vector of the target foreign body data, wherein the bounding box is the position of the target in the image, the category label is the type information of the detected target, and the appearance feature vector is the feature extraction of the target bounding box using a convolutional neural network and the generation of the appearance feature vector.

6. The method for monitoring power transmission lines against external damage based on deep learning target detection and tracking according to claim 5, characterized in that: In S4, the target foreign body data is tracked using the DeepSORT model, which also includes: The bounding box and appearance feature vector of the target foreign body data are input into the standard DeepSORT model. The standard DeepSORT model uses a Kalman filter to predict the motion state of the target, where the motion state includes position, speed and direction; The standard DeepSORT model then matches the currently detected target with the already existing targets based on the similarity metric; The trajectory is updated according to the matching results, wherein the trajectory is updated using the update step of the Carl filter for the successfully matched target; for the unmatched detection results, a new trajectory is initialized; for the unmatched trajectory, the matching count is increased, and the trajectory is deleted after exceeding the preset number; After the trajectory update is completed, the target tracking result in each frame of the target foreign object data is input, and the target tracking result includes the position, speed and direction of the foreign object; Finally, the target tracking results are labeled as standard tracking data.

7. The power transmission line anti-external damage monitoring method based on deep learning target detection and tracking according to claim 6 is characterized in that: Perform anomaly analysis on the standard tracking data in S5 and judge abnormal events based on the anomaly analysis results, including: First, define abnormal events into abnormal indicators, including speed abnormality, trajectory deviation, stagnation detection and target loss; Standard tracking data is analyzed for anomalies using real-time monitoring, model analysis, and cluster analysis; Real-time monitoring is to monitor each target through defined abnormal indicators in each frame of standard tracking data; model analysis is to use deep learning models to judge standard tracking data; cluster analysis is to use clustering algorithms to analyze the movement pattern of the target; The three abnormal analysis results are fused to obtain abnormal events in the standard tracking data, and the abnormal events are marked as transmission line monitoring data.

8. The power transmission line anti-external damage monitoring method based on deep learning target detection and tracking according to claim 7 is characterized in that: In S6, an abnormal event in the transmission line monitoring data is pre-warned and judged, an alarm is issued according to the judgment result, and monitoring reports are generated for the alarm data and the transmission line monitoring data respectively and transmitted to the display terminal, including: Conduct abnormal event analysis on the transmission line monitoring data, including analyzing the specific type, frequency and duration of abnormal events; The analyzed abnormal events are divided into severity according to the degree of abnormality, and the severity is divided into slight abnormality, moderate abnormality and severe abnormality; The warning judgment is made according to the severity of the classification. When the threshold of the severity of the classification is triggered to the warning threshold, the warning alarm is generated. Different severities correspond to different degrees of warning. Visualize the analyzed abnormal events and generated warning alerts using visualization tools; After visualization processing, the data is transmitted to the display terminal for report display.

Images