A distribution network drone infrared inspection method and device based on edge box

Abstract

The present application provides a method and device for infrared inspection of distribution network by drone based on edge box, including: receiving the path planning adjustment instruction transmitted by cloud server and performing adaptive path planning to obtain the inspection path; controlling the drone to perform real-time infrared imaging of distribution network lines and associated support structures according to the inspection path to obtain infrared image data; applying artificial intelligence algorithm model to identify the components of power equipment according to infrared image data, and applying YOLOv8 model to identify the temperature information of each component; sending the components and corresponding temperature information to cloud server to determine whether there is temperature abnormality and generate inspection report. Adaptive path planning ensures that effective inspection can be completed in the shortest time, and the deep integration of drone inspection and edge computing reduces the pressure of server data processing, shortens the time interval from data collection to decision support, and effectively ensures the safety and stability of power grid operation.

Description

Technical Field

[0001] The present application belongs to the field of drone inspection technology, and in particular, relates to a distribution network drone infrared inspection method and device based on an edge box. Background Art

[0002] With the continuous expansion of the scale and complexity of the power system, the maintenance and management of the distribution network has become increasingly important. With the development of drone technology, drones have been widely used in the inspection of power distribution network line towers. However, although the existing drone inspection technology has improved the inspection efficiency, it still has shortcomings in real-time and intelligent processing. Especially in infrared detection, how to quickly and accurately identify potential thermal fault points has become a new challenge. In addition, due to the wide distribution of distribution network line towers, if all the large amount of data generated by drone inspections is transmitted back to the central server for processing, it will not only consume a large amount of graphics processing unit (GPU) resources and network resources, but may also affect the data processing efficiency due to problems such as insufficient server GPU resources. Therefore, there is an urgent need for an intelligent processing method that can improve the inspection data processing efficiency and accurately identify thermal defects. Summary of the invention

[0003] The main purpose of the embodiments of the present invention is to provide a distribution network drone infrared inspection method and device based on an edge box. Adaptive path planning ensures that effective inspection can be completed in the shortest time, deeply integrates drone inspection technology and edge computing concepts, reduces the data processing pressure of the server, shortens the time interval from data collection to decision support, and effectively ensures the safety and stability of power grid operation.

[0004] In a first aspect, a distribution network drone infrared inspection method based on an edge box is provided, the method comprising: receiving a path planning adjustment instruction transmitted by a cloud server and performing adaptive path planning according to the path planning adjustment instruction to obtain an adjusted inspection path, the path planning adjustment instruction is generated by the cloud server according to real-time feedback temperature information and historical data, and the path planning adjustment instruction includes the type and priority of the current inspection task; controlling the drone to perform real-time infrared imaging of the distribution network line and the associated support structure according to the inspection path to obtain infrared image data; applying an artificial intelligence algorithm model according to the infrared image data to identify the components of the power equipment, and applying a YOLOv8 model to identify the temperature information of each component, wherein the infrared image data is preprocessed based on the temperature attribute of the infrared image data; applying a multi-scale feature fusion module to perform feature fusion on components of different sizes according to the preprocessed infrared image data to obtain a feature map; applying a temperature-aware feature enhancement module to perform temperature-sensitive detection on the feature map based on the following loss function to obtain enhanced feature information: ,in, The classification loss is used to measure the gap between the predicted category and the true category. The bounding box position loss is used to measure the position deviation between the predicted bounding box and the true bounding box. is the object existence loss, is the temperature sensitive loss, , , is the weight coefficient; the temperature perception post-processing module is used to process the feature information output by the temperature perception feature enhancement module to obtain the temperature information of each component; the temperature sensitivity loss is the sum of the squares of the difference between the actual temperature and the predicted temperature of each pixel in all feature maps:

[0005]

[0006] in, is the number of feature maps, is the height of the feature map, is the width of the feature map, is the true temperature, It predicts the temperature; sends the components and corresponding temperature information to the cloud server to determine whether there is a temperature anomaly and generate an inspection report.

[0007] In one possible implementation, adaptive path planning is performed according to the path planning adjustment instruction to obtain an adjusted inspection path, including: if the type of the current inspection task is a troubleshooting task or an emergency response task, the inspection path is adjusted so that the planned drone preferentially passes through the areas corresponding to the troubleshooting task or the emergency response task; if the type of the current inspection task is a regular inspection task, the inspection path is adjusted so that the planned drone skips or quickly passes through the areas corresponding to the regular inspection tasks.

[0008] In another possible implementation, adaptive path planning is performed according to the path planning adjustment instruction to obtain the adjusted inspection path, and further includes: Obtain environmental information, including comprehensive weather conditions and topography; adjust inspection routes based on environmental information.

[0009] In another possible implementation, preprocessing the infrared image data based on the temperature attribute of the infrared image data includes: normalizing the infrared image data by applying the following relationship based on the temperature attribute of the infrared image data:

[0010]

[0011] in, is the temperature correction factor, I is the original infrared image data, is the normalized infrared image data, and represent the mean and standard deviation respectively; Gaussian filtering and / or median filtering are performed on the normalized infrared image data to obtain preprocessed infrared image data.

[0012] In another possible implementation, after the components and the corresponding temperature information are sent to the cloud server, it includes: the cloud server matches the components and the corresponding temperature information with the distribution network asset ledger database, compares with the preset temperature standard, determines whether the temperature of each component is normal, and generates a defect report if there is a temperature abnormality; and displays the processed infrared image data and defect report through a full-featured and user-friendly Web interface.

[0013] In the second aspect, a distribution network drone infrared inspection device based on an edge box is provided, and the device includes: a path adjustment unit, which is used to receive a path planning adjustment instruction transmitted by a cloud server and perform adaptive path planning according to the path planning adjustment instruction to obtain an adjusted inspection path, and the path planning adjustment instruction is generated by the cloud server according to real-time feedback temperature information and historical data, and the path planning adjustment instruction includes the type and priority of the current inspection task; a data acquisition module, which is used to receive infrared image data obtained by the drone performing infrared imaging of the distribution network line and the associated support structure; a component temperature recognition module, which is used to identify the components of the power equipment according to the infrared image data by applying an artificial intelligence algorithm model, and at the same time, applying the YOLOv8 model to identify the temperature information of each component, wherein the infrared image data is preprocessed based on the temperature attribute of the infrared image data; according to the preprocessed infrared image data, a multi-scale feature fusion module is applied to perform feature fusion on components of different sizes to obtain a feature map; and a temperature-aware feature enhancement module is applied to perform temperature-sensitive detection on the feature map based on the following loss function to obtain enhanced feature information: ,in, The classification loss is used to measure the gap between the predicted category and the true category. The bounding box position loss is used to measure the position deviation between the predicted bounding box and the true bounding box. is the object existence loss, is the temperature sensitive loss, , , is the weight coefficient; the temperature perception post-processing module is used to process the feature information output by the temperature perception feature enhancement module to obtain the temperature information of each component; the temperature sensitivity loss is the sum of the squares of the difference between the actual temperature and the predicted temperature of each pixel in all feature maps:

[0014]

[0015] in, is the number of feature maps, is the height of the feature map, is the width of the feature map, is the true temperature, It is the predicted temperature; the data sending module is used to send the component parts and the corresponding temperature information to the cloud server to determine whether there is a temperature anomaly and generate an inspection report.

[0016] In a third aspect, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, an edge box-based distribution network drone infrared inspection method as provided in the first aspect is implemented.

[0017] In a fourth aspect, a non-transitory computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, an edge box-based distribution network drone infrared inspection method as provided in the first aspect is implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in describing the embodiments of the present application are briefly introduced below.

[0019] Figure 1 A schematic diagram of the structure of a distribution network drone infrared inspection system based on an edge box provided by an embodiment of the present invention;

[0020] Figure 2 A flow chart of a method for infrared inspection of a distribution network by a drone based on an edge box provided by an embodiment of the present invention;

[0021] Figure 3 A structural diagram of a distribution network drone infrared inspection device based on an edge box provided by an embodiment of the present invention;

[0022] Figure 4 A schematic diagram of the physical structure of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar modules or modules with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and cannot be interpreted as limiting the present invention.

[0024] It will be understood by those skilled in the art that, unless expressly stated, the singular forms "a", "an", "said" and "the" used herein may also include plural forms. It should be further understood that the term "comprising" used in the specification of the present application refers to the presence of the features, integers, steps, operations, modules and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, modules, components and / or groups thereof. It should be understood that when we refer to a module as being "connected" or "coupled" to another module, it may be directly connected or coupled to the other modules, or there may be intermediate modules. In addition, the "connection" or "coupling" used herein may include wireless connection or wireless coupling. The term "and / or" used herein includes all or any modules and all combinations of one or more associated listed items.

[0025] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation method of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0026] The technical solution of the present application and how to solve the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0027] like Figure 1 The edge box-based distribution network drone infrared inspection system provided in the embodiment of the present invention includes: a drone, an edge box equipped on the drone, and a cloud server connected to the edge box in 4G communication.

[0028] The drone is equipped with a high-precision infrared camera and a visible light camera. The drone takes infrared images of the distribution network lines and associated supporting structures to obtain infrared image data. Specifically, the drone takes off and collects data. The drone flies along a preset route and uses an infrared camera and a visible light camera to take photos of the distribution network line towers.

[0029] The edge box is an edge computing device, which can be an electronic device such as a computer or processor. The infrared images taken by the drone are efficiently cached in the edge box through the built-in SDK protocol and customized file transfer protocol of the edge box. In order to ensure the stability and security of the data transmission process and prevent data loss or damage due to signal interference or interruption, a reliable data transmission protocol is adopted, and a stable connection is established between the edge box and the drone to ensure the safe transmission of data.

[0030] The edge box is equipped with a graphics processing unit (GPU) and a central processing unit (CPU). The graphics processing unit GPU is used to execute AI recognition algorithms and handle parallel computing tasks. The central processing unit CPU is used to execute temperature analysis algorithms and handle sequential computing tasks. The edge box uses artificial intelligence algorithm models to identify the components of power equipment based on infrared image data, and obtains the temperature information corresponding to each component. Once an abnormal temperature is found, a preliminary report with the tower number and temperature information is immediately generated. Through the 4G communication module, the processed data (including component identification results and temperature information) is transmitted to the cloud server in real time. The 4G communication module uses HTTP or HTTPS protocols and uses encryption to ensure data security. The processed data is encoded into a data packet that complies with the protocol and sent to the cloud server via a wireless network. In this way, the accuracy of AI recognition and the accuracy of temperature measurement are improved, so that potential overheating problems can be discovered in a timely manner.

[0031] The embodiment of the present invention separates the tasks of the AI ​​recognition algorithm and the temperature analysis algorithm, and runs them on the GPU and CPU respectively, so as to give full play to the advantages of their respective hardware. Specifically, the AI ​​recognition algorithm is arranged to be executed on the GPU of the edge box because it needs to process a large number of parallel computing tasks. This not only effectively utilizes the powerful parallel processing capabilities of the GPU, but also significantly improves the speed and accuracy of image recognition. At the same time, the temperature analysis algorithm is handled by the CPU, considering that the CPU is more efficient in processing sequential computing tasks. This design not only optimizes the use of hardware resources, but also greatly reduces the cost of the central server, because it is no longer necessary to equip the server with an expensive GPU, allowing the server to focus on tasks such as data management and report generation.

[0032] The cloud server is used to cache infrared images taken by drones, and is deployed with an AI algorithm model to process the cached infrared images in real time; wherein, the AI ​​algorithm model includes but is not limited to a convolutional neural network (CNN) or a deep learning model. After receiving the data, the server backend matches it with the distribution network asset ledger database and combines it with the temperature matrix for in-depth statistical analysis and pattern recognition. According to the preset temperature standard, it automatically determines whether there is a risk of overheating or thermal failure, and generates a defect report accordingly. Specifically, the cloud server analyzes the inspection status of each component based on the components and the corresponding temperature information combined with the distribution network asset ledger database, and generates and displays the inspection report. A comprehensive and user-friendly web interface is developed on the cloud server, so that operation and maintenance personnel can access and view infrared image data and defect reports in real time through the Internet. The web interface design is intuitive and easy to use, and provides a wealth of visualization tools to help users understand the meaning behind the data. The web interface supports real-time display of infrared image data, and can associate it with the distribution network asset ledger database according to the input line name to find the corresponding tower longitude and latitude information, and locate the infrared image to the specific tower position based on the longitude and latitude information; the positioning accuracy is not less than 0.1 meters.

[0033] After receiving the data from the edge box, the cloud server evaluates the priority of the current inspection task in combination with the historical data in the distribution network asset ledger database. Historical data includes previous inspection records, equipment maintenance records, etc., which helps to more accurately determine which areas or equipment need to be focused on. Based on the real-time feedback of temperature information, combined with historical data and preset rules in the distribution network asset ledger database, the priority of the current inspection task is evaluated to dynamically adjust the inspection path of the drone. The cloud server sends a path planning adjustment instruction to the drone to adjust its inspection path. For example:

[0034] Troubleshooting: If the components of a zone show signs of overheating, give that zone a priority for detailed inspection;

[0035] Emergency response: If an emergency occurs (such as a natural disaster or man-made damage), the system will immediately activate the emergency response mode and quickly dispatch drones to the affected area for rapid assessment;

[0036] Routine inspection: If there is no obvious problem with a certain line section, it can be quickly skipped to avoid unnecessary repeated inspections, thereby improving overall inspection efficiency.

[0037] Consider multiple factors: Path planning not only considers temperature anomalies, but also weather conditions, topography and other factors to ensure that the drone can complete the inspection task safely and efficiently. For example, in severe weather conditions, the system may recommend postponing inspections in certain high-risk areas to avoid unnecessary risks.

[0038] Routine inspections are regular inspections of distribution network lines and associated towers to ensure the normal operation of equipment. Multi-area routine inspections are regular comprehensive inspections of multiple distribution network areas to ensure the normal operation of equipment in each area. The system automatically plans the inspection route of the drone based on the preset schedule and historical data, focusing on the inspection of component equipment and areas prone to problems. Through the lightweight AI algorithm model, infrared image data is analyzed in real time to identify potential problems and generate reports.

[0039] Troubleshooting means that when a fault alarm occurs on a certain line or device, the system immediately starts the troubleshooting task. The drone quickly goes to the fault area, uses the AI ​​algorithm on the edge box to analyze the temperature anomaly in real time, and generates a detailed fault report. Multi-area troubleshooting means that when a fault alarm occurs on a certain line or device in multiple areas, the system immediately starts the troubleshooting task. Multiple drones quickly go to the fault area, use the AI ​​algorithm on the edge box to analyze the temperature anomaly in real time, and generate a detailed fault report. The cloud server combines historical data and preset rules to guide operation and maintenance personnel to take the most effective repair measures.

[0040] Emergency response is when a natural disaster or emergency occurs, the system quickly activates the emergency response mode to assess the situation in the affected area. Drones are immediately dispatched to the disaster area for comprehensive infrared imaging and data analysis. Multi-region emergency response is when a natural disaster or emergency occurs, multiple drones are immediately dispatched to the disaster area for comprehensive infrared imaging and data analysis. The system uses a real-time feedback mechanism to send the latest disaster information to the command center in a timely manner, helping decision makers to respond quickly.

[0041] In the implementation of the present invention, the system can also intelligently allocate drones and other inspection resources according to the requirements of different tasks. For example, in large-scale troubleshooting, multiple drones can be dispatched at the same time to work together to speed up the inspection. Adaptive path planning enables drones to complete effective inspections in the shortest time, reducing energy consumption and maintenance costs.

[0042] An efficient information transmission mechanism is realized through the 4G communication module, which not only improves the data transmission efficiency, but also simplifies the user's operation process. A simple and easy-to-use web interface has been developed on the front end, through which users can view the processed infrared data images and the temperature conditions of related equipment in real time. It enables operation and maintenance personnel to access and view the processed infrared image data and corresponding defect reports in real time through the Internet. The interface supports a variety of visualization tools to help users better understand the meaning behind the data. This instant feedback mechanism helps to quickly identify and respond to possible thermal failure risks, so that operation and maintenance personnel can understand the inspection situation at the first time, take timely measures to deal with potential risks, and shorten the time interval from data collection to decision support, thereby ensuring the safety and stability of power grid operation.

[0043] The real-time feedback and adaptive inspection path optimization system based on edge boxes can not only efficiently handle various types of inspection tasks, but also flexibly adjust strategies according to actual conditions to ensure the efficiency and pertinence of inspection work and improve the safety and stability of power grid operation. When handling inspection tasks in multiple areas, the real-time feedback and adaptive inspection path optimization system based on edge boxes adopts intelligent task allocation, dynamic path planning, and multi-UAV collaborative operation mechanisms to ensure efficient and flexible completion of inspection work.

[0044] Over time, the system will accumulate a large amount of inspection data, including temperature distribution characteristics under different seasons and weather conditions. By learning from these data, the system can continuously optimize the inspection strategy. Machine learning algorithms (such as decision trees, random forests, neural networks, etc.) can be used to extract features and recognize patterns from inspection data to predict possible fault points. For example, the system can identify equipment or areas that are prone to problems in advance by learning historical data, so as to formulate more targeted inspection plans. Based on the accumulated data and learning results, the path optimization algorithm can continue to evolve as the amount of data grows and provide more accurate services. For example, the system can plan more efficient inspection paths in advance, reduce the number of unnecessary inspections, and further ensure the safe and stable operation of the power grid. The path optimization algorithm based on machine learning can continue to evolve as the amount of data grows, provide more accurate services, and adapt to complex and changing actual application scenarios. The intelligent learning and optimization mechanism ensures the safety and reliability of the inspection path, reduces the possibility of human operation errors, and further improves the stability of the entire system.

[0045] The embodiment of the present invention deeply integrates the drone inspection technology and edge computing concepts. By using a specific software development kit (SDK) protocol for communication, the edge box can seamlessly connect with the sensor on the drone and quickly obtain the infrared image data taken by the drone. These data can be processed instantly on the edge side without uploading to the cloud or central server, thus avoiding transmission delays caused by network instability and reducing the data processing pressure on the central server.

[0046] like Figure 2 The flowchart of a method for infrared inspection of a distribution network by a drone based on an edge box provided by an embodiment of the present invention is shown, which is applied to an edge box equipped on a drone. The method includes:

[0047] Step S11, receive the path planning adjustment instruction transmitted by the cloud server and perform adaptive path planning according to the path planning adjustment instruction to obtain the adjusted inspection path. The path planning adjustment instruction is generated by the cloud server based on real-time feedback of temperature information and historical data. The path planning adjustment instruction includes the type and priority of the current inspection task.

[0048] Optionally, the higher the priority, the inspection path is adjusted to preferentially pass through the areas corresponding to high-priority or more urgent tasks. If the current inspection task type is a troubleshooting task or an emergency response task, the inspection path is adjusted so that the planned drone preferentially passes through the areas corresponding to the troubleshooting task or the emergency response task; if the current inspection task type is a regular inspection task, the inspection path is adjusted so that the planned drone skips or quickly passes through the areas corresponding to the regular inspection tasks.

[0049] The path planning adjustment instructions are generated by the cloud server based on the real-time feedback of temperature information and historical data. The path planning adjustment instructions include the type and priority of the current inspection task. When a fault alarm occurs in a certain line or equipment, the troubleshooting task and the corresponding priority are set. The troubleshooting task is immediately started, and at least one drone is controlled to quickly go to the fault area for inspection and generate a detailed fault report. In the event of a natural disaster or emergency, the emergency response task and the corresponding priority are set. The emergency response task is immediately started, and at least one drone is controlled to go to the corresponding area for inspection and assess the situation in the affected area. Through the real-time feedback mechanism, the latest disaster information is sent to the command center in a timely manner to help decision makers respond quickly.

[0050] In the embodiment of the present invention, environmental information may also be obtained, the environmental information including comprehensive weather conditions and topography; and the inspection path may be adjusted according to the environmental information.

[0051] In addition, as time goes by, a large amount of inspection data will be accumulated, including temperature distribution characteristics under different seasons and weather conditions. Through the cloud server's learning of these data, the system can continuously optimize the inspection strategy. Machine learning algorithms (such as decision trees, random forests, neural networks, etc.) can be used to extract features and recognize patterns from inspection data to predict possible fault points. The path planning adjustment instructions can also include equipment or areas that are prone to problems. Based on the equipment or areas that are prone to problems, adaptive path planning is performed to obtain the adjusted inspection path. In this way, a more targeted and efficient inspection path can be formulated to reduce the number of unnecessary inspections. In this way, as the amount of data continues to grow, more accurate services can be provided to further ensure the safe and stable operation of the power grid.

[0052] Step S12, controlling the UAV to perform real-time infrared imaging of the distribution network lines and associated supporting structures according to the inspection path to obtain infrared image data.

[0053] Optionally, the drone can be controlled to fly autonomously along the inspection path without human intervention; the high-precision infrared camera on the drone can be used to capture thermal images of the distribution network lines and associated towers in real time. To ensure that the image quality and coverage meet the inspection requirements, the flight path and shooting angle of the drone are pre-planned and adjusted appropriately according to the real-time environmental conditions during the flight.

[0054] Step S13, using an artificial intelligence algorithm model to identify components of the power equipment based on the infrared image data, and using a YOLOv8 model to identify temperature information of each component.

[0055] A lightweight AI algorithm model is deployed in the edge box to process cached infrared images in real time. The AI ​​algorithm model can accurately identify the components of power equipment and analyze their surface temperature distribution. In this way, the accuracy of AI recognition and the precision of temperature measurement are improved, so that potential overheating problems can be discovered in a timely manner.

[0056] In the embodiment of the present invention, the following formula is used to measure temperature:

[0057]

[0058] in, It is in pixels The average temperature at It is in The temperature reading at the pixel during the first measurement is is the number of measurements.

[0059] In step S12, optionally, an artificial intelligence algorithm model is applied to identify components of the power equipment based on the infrared image data; and a YOLOv8 model is applied to identify temperature information of each component based on the preprocessed infrared image data. The components include but are not limited to transformers, circuit breakers, tension clamps, etc.

[0060] The process of applying the YOLOv8 model to process the preprocessed infrared image data includes: preprocessing the infrared image data based on the temperature attribute of the infrared image data; applying the multi-scale feature fusion module to perform feature fusion on components of different sizes according to the preprocessed infrared image data to obtain a feature map; applying the temperature-aware feature enhancement module to perform temperature-sensitive detection on the feature map to obtain enhanced feature information; applying the temperature-aware post-processing module to process the feature information output by the temperature-aware feature enhancement module to obtain the temperature information of each component.

[0061] In the embodiment of the present invention, a temperature correction factor is introduced during the input preprocessing stage. , to take into account the temperature properties unique to infrared images. Based on the temperature properties of infrared image data, the following relationship is applied to normalize the infrared image data:

[0062]

[0063] in, is the temperature correction factor, I is the original infrared image data, that is, the original infrared image data without cropping, scaling or normalization, is the normalized infrared image data, and represent mean and standard deviation respectively. The temperature correction factor can be dynamically adjusted according to the temperature range of different scenarios. The introduction of is to compensate for the measurement error caused by changes in ambient temperature and ensure that consistent normalization results can be obtained under different temperature conditions.

[0064] Then, the normalized infrared image data is subjected to Gaussian filtering and / or median filtering to obtain preprocessed infrared image data. The noise interference in the infrared image is reduced by introducing a noise suppression module. The noise suppression formula is: ,in is a denoising function, such as Gaussian filtering or median filtering. The function of Gaussian filtering and median filtering is to remove random noise from the image while retaining the important features of the image as much as possible. Gaussian filtering is a linear smoothing filter that is suitable for removing Gaussian noise and can effectively smooth the image. The median filter is particularly effective in removing salt and pepper noise because it replaces the value of the central pixel by taking the median value in the neighborhood, which can well protect the edge information from being affected.

[0065] The combination of normalization and noise suppression is to improve the quality of infrared images and make them more suitable for subsequent analysis or processing tasks. Temperature correction ensures the consistency and accuracy of the data, while noise suppression improves the clarity and usability of the image.

[0066] In the forward propagation process of the YOLOv8 model, a multi-scale feature fusion module (MSFF) is added to enhance the model's ability to detect objects of different sizes. The feature map F is generated by this module, that is, F = MSFF ). At the same time, in order to further improve the detection capability of temperature-sensitive targets, we also introduced the Temperature-Aware Feature Enhancement (TA-FE) module. The formula for temperature-aware feature enhancement is: ,in, It is the output feature map after being processed by the temperature perception feature enhancement module. is the input feature map, is the feature map generated by the multi-scale feature fusion module, is the temperature information of the input image. The structure of the output tensor O has also changed, and its shape is now ,in, is the batch size, that is, the number of feature maps in a batch, indicating the number of images processed in one forward propagation. is the height of the feature map, is the width of the feature map, is the basic feature dimension, which is the number of basic features predicted for each grid unit, and 5 represents the basic prediction information for each target, including 4 coordinates and 1 confidence score, is an additional feature dimension used to represent temperature information, is the output dimension of the temperature-aware feature enhancement module, representing the enhanced feature information. The new output tensor is generated by the YOLOv8 model, namely .

[0067] By adding a multi-scale feature fusion mechanism to the traditional YOLOv8, the model can better handle targets of various sizes, especially in small target detection. The application of the temperature-aware feature enhancement module can improve detection performance in temperature-sensitive scenarios.

[0068] In the embodiment of the present invention, the temperature-aware feature enhancement module is applied to perform temperature-sensitive detection on the feature map based on the following loss function to obtain enhanced feature information:

[0069]

[0070] in, The classification loss is used to measure the gap between the predicted category and the true category. The bounding box position loss is used to measure the position deviation between the predicted bounding box and the true bounding box. is the object existence loss, is the temperature sensitive loss, , , is the weight coefficient.

[0071] That is, in the loss calculation of the YOLOv8 model, an adaptive loss function is designed and adaptive weights are introduced. and , these two parameters will dynamically adjust the classification loss according to the performance during training and bounding box location loss The total loss formula thus becomes In addition, temperature-sensitive losses are introduced , in order to further improve the model's ability to detect temperature-sensitive targets. The temperature-sensitive loss formula is: , which is the sum of the squares of the difference between the actual temperature and the predicted temperature of each pixel in all feature maps, where is the true temperature, is the predicted temperature. By minimizing this loss term, the model will strive to make the predicted temperature as close to the actual temperature as possible, thereby improving the detection ability of temperature-sensitive targets. , so that the model not only focuses on the classification and location of the target, but also learns the temperature characteristics of the target, which is particularly important for target detection in infrared imaging.

[0072] Combining all loss terms, the total loss formula is ,in is the weight of the temperature-sensitive loss, which is initially 1 and can be dynamically adjusted according to the performance during training.

[0073] Through adaptive weight and , the model can automatically adjust the importance of classification loss and bounding box position loss during training to avoid excessive loss of one type affecting the overall effect. It can be dynamically adjusted according to the performance of each part during training to achieve better overall performance. The total loss formula comprehensively considers the losses of classification, position and temperature to ensure that the model can achieve good performance in all aspects.

[0074] In the post-processing of the YOLOv8 model, an improved non-maximum suppression (NMS) method is proposed, namely, temperature-sensitive NMS (TS-NMS), which can dynamically adjust the temperature sensitivity threshold according to the temperature distribution of the infrared image. , thereby more effectively screening out the most likely target instances, especially when facing complex backgrounds. Traditional NMS methods usually decide whether to retain a detection box based on the degree of overlap of bounding boxes (such as IoU, Intersection over Union). Temperature-sensitive non-maximum suppression (TS-NMS) adds a temperature factor on this basis, and can dynamically adjust its threshold according to the temperature distribution of the target in the infrared image. This means that in areas with large temperature changes, TS-NMS can more flexibly screen out real targets and reduce false detections or missed detections.

[0075] In addition, a temperature-aware post-processing module (TA-PP) is introduced to further optimize the detection results. The temperature-aware post-processing formula is: The temperature-aware post-processing module (TA-PP) further optimizes the detection results by combining temperature information and other features to improve the quality of the final output. The temperature-aware post-processing formula shows that the original detection results and temperature information is input into this module and processed to obtain optimized output This step helps to better understand the relationship between the target and the background, especially when the temperature difference is significant, which can effectively improve the detection accuracy. In this way, the temperature information is further used to optimize the detection results in the post-processing stage, which improves the accuracy and reliability of target detection in complex backgrounds.

[0076] In this way, the output of the model not only includes the traditional bounding box, category, and confidence score, but also adds the temperature information of the target. The output format is ,in are the center coordinates of the target bounding box, are the width and height of the target bounding box; is the object confidence score, which indicates the possibility of the existence of the object in the bounding box; is the probability score of the target belonging to each category; is the temperature information of the target, It is the output feature of the temperature-aware feature enhancement module.

[0077] The output format contains the usual object bounding box ( ), category, and confidence score, it also provides additional temperature information of the target The output features of the temperature-aware feature enhancement module are enriched by this improvement, which is very useful for application scenarios that require temperature information.

[0078] Through the above changes to the YOLOv8 model, not only the performance of YOLOv8 in processing infrared images is improved, but also its practical value in specific application fields is increased.

[0079] In the embodiment of the present invention, the lightweight AI algorithm deployed in the edge box adopts a lightweight deep learning model, such as MobileNet, SqueezeNet, etc., to ensure that the edge box can process data quickly. Data is processed on the edge box, avoiding network delays and achieving true real-time response.

[0080] In order to handle various types of inspection tasks, multi-level task management and intelligent decision-making mechanisms can also be designed, including task classification and priority setting, dynamic task scheduling, environmental condition assessment, and resource allocation optimization. Not only temperature anomalies are considered, but other factors are also integrated to ensure that drones complete inspection tasks safely and efficiently. In addition, based on the ability of learning and continuous improvement, the accuracy of inspection strategies is continuously improved through data accumulation and feature extraction, as well as the application of machine learning algorithms. For inspection tasks in multiple areas, intelligent task allocation, dynamic path planning, and multi-drone collaborative operations are adopted to ensure efficient and flexible completion of inspection work, whether in routine inspections, troubleshooting, or emergency response scenarios. Excellent performance can be demonstrated.

[0081] Step S14, sending the components and corresponding temperature information to the cloud server to determine whether there is a temperature anomaly and generate an inspection report.

[0082] The components and corresponding temperature information obtained in step S13 are sent to the cloud server via 4G communication. The data transmission adopts HTTP or HTTPS protocol, and the data security is guaranteed by encryption. The processed data is encoded into a data packet that complies with the HTTP or HTTPS protocol and sent to the cloud server via a wireless network.

[0083] After the components and their corresponding temperature information are sent to the cloud server, the cloud server matches the components and their corresponding temperature information with the distribution network asset ledger database, and compares them with the preset temperature standards to determine whether the temperature of each component is normal. If there is a temperature abnormality, a defect report is generated; the processed infrared image data and defect report are displayed through a full-featured and user-friendly Web interface.

[0084] In step S13, after receiving the data, the cloud server backend matches it with the distribution network asset ledger database and combines it with the temperature matrix for in-depth statistical analysis and pattern recognition; based on the preset temperature standard, it automatically determines whether there is a risk of overheating or thermal failure, and generates a defect report accordingly. The temperature standard is defined by the following formula:

[0085]

[0086] If the temperature of any component device presents a risk of overheating or thermal failure, it is flagged as a potential failure point.

[0087] A comprehensive and user-friendly web interface is developed on the cloud server, so that operation and maintenance personnel can access and view the processed infrared image data and corresponding defect reports in real time through the Internet; the web interface includes but is not limited to visualization tools such as infrared image display area, temperature distribution map, defect report list, etc., and supports users to enter the line name and associate it with the ledger system to find the corresponding tower longitude and latitude information, and locate the infrared image to the specific tower position based on the longitude and latitude information; the positioning accuracy is not less than 0.1 meter.

[0088] The cloud server can also remotely monitor the status of the edge box and support remote updates of the AI ​​algorithm model in the edge box.

[0089] In an embodiment of the present invention, the function of real-time monitoring and automatic report generation is realized through a 4G communication module. When the drone carries an edge box to inspect the distribution network area, the infrared images taken are quickly processed and analyzed. Subsequently, the processed data is uploaded to the cloud server in real time through the 4G network. After receiving the data, the back end of the cloud server will automatically perform further analysis and automatically generate a detailed inspection report. At the same time, in order to facilitate users to understand the inspection situation at any time, a simple and easy-to-use Web interface has been developed at the front end, through which users can view the processed infrared data images and the temperature of related equipment in real time. In this way, both on-site engineers and remote managers can obtain the latest inspection results in real time, greatly shortening the time interval from data collection to decision support, and enhancing the responsiveness and practicality of the system. In addition, this instant feedback mechanism also helps to quickly identify and respond to possible thermal failure risks, thereby ensuring the safety and stability of the distribution network operation.

[0090] In summary, the embodiment of the present invention receives the path planning adjustment instruction transmitted by the cloud server and performs adaptive path planning according to the path planning adjustment instruction to obtain an adjusted inspection path, the path planning adjustment instruction is generated by the cloud server according to the real-time feedback temperature information and historical data, and the path planning adjustment instruction includes the type and priority of the current inspection task; the drone is controlled to perform real-time infrared imaging of the distribution network lines and associated support structures according to the inspection path to obtain infrared image data; the artificial intelligence algorithm model is applied according to the infrared image data to identify the components of the power equipment, and the YOLOv8 model is applied to identify the temperature information of each component; the components and the corresponding temperature information are sent to the cloud server to determine whether there is a temperature abnormality and generate an inspection report, which deeply integrates the drone inspection technology and the edge computing concept, avoids the transmission delay caused by network instability, reduces the data processing pressure of the server, and greatly shortens the time interval from data collection to decision support, which helps to quickly identify and respond to possible thermal failure risks, thereby ensuring the safety and stability of power grid operation.

[0091] like Figure 3 The figure shows a structural diagram of a distribution network drone infrared inspection device based on an edge box provided by an embodiment of the present invention, the device comprising:

[0092] The path adjustment unit is used to receive the path planning adjustment instructions transmitted by the cloud server and perform adaptive path planning according to the path planning adjustment instructions to obtain the adjusted inspection path. The path planning adjustment instructions are generated by the cloud server based on real-time feedback of temperature information and historical data. The path planning adjustment instructions include the type and priority of the current inspection task.

[0093] The data acquisition module 301 is used to control the UAV to perform infrared imaging of the distribution network lines and associated support structures according to the inspection path to obtain infrared image data;

[0094] The component temperature recognition module 302 is used to identify the components of the power equipment by applying an artificial intelligence algorithm model according to the infrared image data, and to identify the temperature information of each component by applying a YOLOv8 model;

[0095] The data sending module 303 is used to send the components and the corresponding temperature information to the cloud server to determine whether there is a temperature anomaly and generate an inspection report.

[0096] The device of the above embodiment is applied to the corresponding method in the above embodiment, and has the beneficial effects of the corresponding method embodiment, which will not be described in detail here.

[0097] Figure 4 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 4 As shown, the electronic device may include: a processor (processor) 401, a communication interface (Communications Interface) 402, a memory (memory) 403 and a communication bus 404, wherein the processor, the communication interface, and the memory complete mutual communication through the communication bus. The processor can call the logic instructions in the memory to execute a distribution network drone infrared inspection method based on an edge box, the method comprising: receiving a path planning adjustment instruction transmitted by a cloud server and performing adaptive path planning according to the path planning adjustment instruction, obtaining an adjusted inspection path, the path planning adjustment instruction is generated by the cloud server according to real-time feedback temperature information and historical data, and the path planning adjustment instruction includes the type and priority of the current inspection task; controlling the drone to perform real-time infrared imaging of the distribution network line and the associated support structure according to the inspection path to obtain infrared image data; applying an artificial intelligence algorithm model to identify the components of the power equipment according to the infrared image data, and applying a YOLOv8 model to identify the temperature information of each component; sending the components and the corresponding temperature information to the cloud server to determine whether there is a temperature abnormality and generate an inspection report.

[0098] In addition, the logic instructions in the above-mentioned memory can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0099] On the other hand, an embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer can execute a distribution network drone infrared inspection method based on an edge box provided by the above-mentioned method embodiments, the method including: receiving a path planning adjustment instruction transmitted by a cloud server and performing adaptive path planning according to the path planning adjustment instruction to obtain an adjusted inspection path, the path planning adjustment instruction is generated by the cloud server based on real-time feedback temperature information and historical data, and the path planning adjustment instruction includes the type and priority of the current inspection task; controlling the drone to perform real-time infrared imaging of the distribution network lines and associated support structures according to the inspection path to obtain infrared image data; applying an artificial intelligence algorithm model to identify components of the power equipment according to the infrared image data, and applying a YOLOv8 model to identify the temperature information of each component; sending the components and corresponding temperature information to the cloud server to determine whether there is a temperature abnormality and generate an inspection report.

[0100] On the other hand, an embodiment of the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, it is implemented to execute a distribution network drone infrared inspection method based on an edge box provided in the above-mentioned embodiments, the method comprising: receiving a path planning adjustment instruction transmitted by a cloud server and performing adaptive path planning according to the path planning adjustment instruction to obtain an adjusted inspection path, the path planning adjustment instruction is generated by the cloud server based on real-time feedback temperature information and historical data, the path planning adjustment instruction includes the type and priority of the current inspection task; controlling the drone to perform real-time infrared imaging of the distribution network lines and associated support structures according to the inspection path to obtain infrared image data; applying an artificial intelligence algorithm model to identify components of the power equipment according to the infrared image data, and applying a YOLOv8 model to identify the temperature information of each component; sending the components and corresponding temperature information to the cloud server to determine whether there is a temperature abnormality and generate an inspection report.

[0101] It should be understood that, although the steps in the flowchart of the accompanying drawings are displayed in sequence as indicated by the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least a part of the steps in the flowchart of the accompanying drawings may include multiple sub-steps or multiple stages, and these sub-steps or stages are not necessarily executed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be executed in turn or alternately with other steps or at least a part of the sub-steps or stages of other steps.

[0102] The above is only a partial implementation of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A distribution network drone infrared inspection method based on edge box, characterized in that: The method comprises: Receive a path planning adjustment instruction transmitted by a cloud server and perform adaptive path planning according to the path planning adjustment instruction to obtain an adjusted inspection path, wherein the path planning adjustment instruction is generated by the cloud server according to real-time feedback of temperature information and historical data, and the path planning adjustment instruction includes the type and priority of the current inspection task; Controlling the UAV to perform real-time infrared imaging of the distribution network lines and associated supporting structures according to the inspection path to obtain infrared image data; According to the infrared image data, an artificial intelligence algorithm model is applied to identify the components of the power equipment, and the YOLOv8 model is applied to identify the temperature information of each of the components, wherein the infrared image data is preprocessed based on the temperature attribute of the infrared image data; according to the preprocessed infrared image data, a multi-scale feature fusion module is applied to perform feature fusion on components of different sizes to obtain a feature map; and a temperature-aware feature enhancement module is applied to perform temperature-sensitive detection on the feature map based on the following loss function to obtain enhanced feature information: ,in, The classification loss is used to measure the gap between the predicted category and the true category. The bounding box position loss is used to measure the position deviation between the predicted bounding box and the true bounding box. is the object existence loss, is the temperature sensitive loss, , , is a weight coefficient; the temperature sensing post-processing module is used to process the feature information output by the temperature sensing feature enhancement module to obtain the temperature information of each component; the temperature sensitive loss is the sum of the squares of the difference between the actual temperature and the predicted temperature of each pixel in all feature maps: in, is the number of feature maps, is the height of the feature map, is the width of the feature map, is the true temperature, is the predicted temperature; The components and the corresponding temperature information are sent to a cloud server to determine whether there is a temperature anomaly and generate an inspection report.

2. The method according to claim 1, characterized in that The step of performing adaptive path planning according to the path planning adjustment instruction to obtain an adjusted inspection path includes: If the type of the current inspection task is a troubleshooting task or an emergency response task, the inspection path is adjusted so that the planned drone preferentially passes through the area corresponding to the troubleshooting task or the emergency response task; If the type of the current inspection task is a regular inspection task, the inspection path is adjusted to plan the drone to skip or quickly pass through the area corresponding to the regular inspection task.

3. The method according to claim 1, characterized in that The path planning adjustment instruction also includes equipment or areas prone to problems, and the adaptive path planning is performed according to the path planning adjustment instruction to obtain the adjusted inspection path, and further includes: Adaptive path planning is performed based on the equipment or area prone to problems to obtain an adjusted inspection path.

4. The method according to claim 1, characterized in that The step of performing adaptive path planning according to the path planning adjustment instruction to obtain an adjusted inspection path also includes: Acquiring environmental information, including comprehensive weather conditions and topography; The inspection path is adjusted according to the environmental information.

5. The method according to claim 1, characterized in that The preprocessing of the infrared image data based on the temperature attribute of the infrared image data comprises: The infrared image data is normalized by applying the following relationship based on the temperature attribute of the infrared image data: in, is the temperature correction factor, I is the original infrared image data, is the normalized infrared image data, and represent mean and standard deviation respectively; Gaussian filtering and / or median filtering are performed on the normalized infrared image data to obtain preprocessed infrared image data.

6. The method according to claim 1, characterized in that After sending the component and the corresponding temperature information to the cloud server, the method includes: The cloud server matches the components and the corresponding temperature information with the distribution network asset ledger database, compares with the preset temperature standard, determines whether the temperature of each component is normal, and generates a defect report if there is a temperature abnormality; The processed infrared image data and defect reports are displayed through the web interface.

7. A distribution network drone infrared inspection device based on edge box, characterized in that: The device comprises: A path adjustment unit, used to receive a path planning adjustment instruction transmitted by a cloud server and perform adaptive path planning according to the path planning adjustment instruction to obtain an adjusted inspection path, wherein the path planning adjustment instruction is generated by the cloud server according to real-time feedback of temperature information and historical data, and the path planning adjustment instruction includes the type and priority of the current inspection task; A data acquisition module is used to control the UAV to perform infrared imaging of the distribution network lines and associated supporting structures according to the inspection path to obtain infrared image data; A component temperature recognition module is used to identify the components of the power equipment by applying an artificial intelligence algorithm model according to the infrared image data, and to identify the temperature information of each component by applying a YOLOv8 model, wherein the infrared image data is preprocessed based on the temperature attribute of the infrared image data; a multi-scale feature fusion module is applied to perform feature fusion on components of different sizes according to the preprocessed infrared image data to obtain a feature map; and a temperature-aware feature enhancement module is applied to perform temperature-sensitive detection on the feature map based on the following loss function to obtain enhanced feature information: ,in, The classification loss is used to measure the gap between the predicted category and the true category. The bounding box position loss is used to measure the position deviation between the predicted bounding box and the true bounding box. is the object existence loss, is the temperature sensitive loss, , , is a weight coefficient; the temperature sensing post-processing module is used to process the feature information output by the temperature sensing feature enhancement module to obtain the temperature information of each component; the temperature sensitive loss is the sum of the squares of the difference between the actual temperature and the predicted temperature of each pixel in all feature maps: in, is the number of feature maps, is the height of the feature map, is the width of the feature map, is the true temperature, is the predicted temperature; The data sending module is used to send the components and the corresponding temperature information to the cloud server to determine whether there is a temperature anomaly and generate an inspection report.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, an edge box-based UAV infrared inspection method for a distribution network is implemented as described in any one of claims 1-6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, an edge box-based UAV infrared inspection method for a distribution network is implemented as described in any one of claims 1-6.

Images