Composite insulator heat defect type identification method and system, electronic device and medium

Abstract

This invention relates to the field of intelligent power line inspection technology, and particularly to a method, system, electronic device, and medium for identifying the type of thermal defects in composite insulators. The method includes: acquiring infrared image data of a target composite insulator; performing insulator target detection and semantic segmentation to obtain the main outline of the insulator; extracting a heating region from the main outline of the insulator, and determining whether a thermal defect exists based on a comparison of the temperature characteristics of the heating region with a preset threshold; if a thermal defect exists, calculating positional parameters based on the spatial relationship between the heating region and the end region of the insulator; and determining the type of thermal defect based on the positional parameters. This approach solves the problem of insufficient accuracy in identifying the type of thermal defects in composite insulators in existing technologies for UAV inspection scenarios, improving the accuracy of thermal defect identification and the reliability of risk classification.

Description

Technical Field

[0001] This invention relates to the field of intelligent power inspection technology, and in particular to a method, system, electronic device and medium for identifying the type of thermal defects in composite insulators. Background Technology

[0002] With the continuous advancement of smart grid construction, ensuring the safe and stable operation of transmission lines has become a core task for power companies. Composite insulators, with their excellent anti-pollution flashover performance and light weight, have been widely used in transmission lines. However, during long-term operation, composite insulators may experience localized abnormal heating due to material aging, internal defects (such as core rod rot), or external damage. These heating defects pose a significant threat to power grid safety. Currently, using drones equipped with infrared thermal imagers for transmission line inspection has become the mainstream approach. Compared to traditional manual ground inspections or manned helicopter inspections, drone inspections offer significant advantages such as flexible location selection, lack of terrain limitations, and high detection efficiency. This effectively reduces the workload of inspection personnel, expands the inspection coverage, and enables large-scale, routine infrared monitoring of transmission line equipment.

[0003] In terms of data processing, existing technologies have begun to apply computer vision and basic image processing algorithms to automatically analyze infrared images collected by drones to identify composite insulators with abnormal heating. These methods typically determine whether the insulator has an overall abnormal temperature by setting temperature thresholds or simple regional feature analysis, thus realizing a shift from "manual image interpretation" to "automatic screening" to some extent and improving image analysis efficiency. However, existing technical solutions still have significant limitations and cannot meet the urgent needs of refined and intelligent operation and maintenance. For example, prior art document 1 (application publication number CN120976204A) discloses a system, device, and method for judging heating defects in edge-end composite insulators. This system identifies heating defects through target detection, segmentation, and centerline temperature extraction, but the method mainly focuses on the overall judgment of the heating state and does not make refined distinctions between the types of heating defects. Specifically, heating defects in composite insulators can be divided into "end heating" and "non-end heating" according to their location and cause. Among them, non-end heating often indicates internal core rod defects and is far more harmful than end heating. The technical solution disclosed in document 1 can only make a binary judgment on whether or not heat is generated. It lacks precise analysis of the spatial location of the heat source and cannot distinguish the type of heat defect. This makes it difficult for maintenance personnel to classify risks and trace the source based on the test results, which affects the efficiency of the maintenance strategy formulation.

[0004] Furthermore, existing technologies generally employ a "fly-first data acquisition, then data transmission and processing" model. This means that after the drone completes its flight path, it stores infrared image data on a medium and transmits it back to a backend server for batch analysis. This model suffers from significant time delays; the cycle from discovering a potential hazard to confirming a work order can take hours or even days. During this period, high-risk defects may continue to worsen, missing the optimal window for early warning and intervention. Simultaneously, this model cannot support real-time interaction and verification at the inspection site. Drone operators cannot perform multi-angle re-photographing of the target based on real-time feedback, increasing maintenance and time costs. Therefore, existing technologies suffer from insufficient accuracy in identifying the type of thermal defects in composite insulators during drone inspections. Summary of the Invention

[0005] To address the aforementioned shortcomings or deficiencies, this invention provides a method, system, electronic device, and medium for identifying the thermal defect type of composite insulators, which can solve the problem of insufficient accuracy in identifying the thermal defect type of composite insulators in the context of UAV inspection.

[0006] This invention provides a method for identifying the type of thermal defects in composite insulators, comprising: Acquire infrared image data of the target composite insulator.

[0007] Insulator target detection and semantic segmentation are performed based on infrared image data to obtain the main outline of the insulator.

[0008] The heating area is extracted from the outline of the insulator body, and the presence of heating defects is determined by comparing the temperature characteristics of the heating area with a preset threshold.

[0009] If there are heating defects in the main outline of the insulator, the position-related parameters are calculated by the cross-comparison algorithm based on the spatial relationship between the heating area and the end area of ​​the insulator.

[0010] The type of heat defect is determined based on location-related parameters and preset heat defect classification configuration. The heat defect types include end heat and non-end heat.

[0011] The intersection-to-intersection ratio algorithm is configured to calculate the ratio of the intersection area of ​​the heating region and the insulator end region to the area of ​​the heating region.

[0012] According to a second aspect, the present invention provides a system for identifying the type of thermal defects in composite insulators, comprising: The infrared image data acquisition module is used to acquire infrared image data of the target composite insulator.

[0013] The insulator body contour segmentation module is used to perform insulator target detection and semantic segmentation operations based on infrared image data to obtain the insulator body contour.

[0014] The heating defect judgment module is used to extract the heating area from the outline of the insulator body and judge whether there is a heating defect based on the comparison between the temperature characteristics of the heating area and a preset threshold.

[0015] The position-related parameter calculation module is used to calculate position-related parameters based on the spatial relationship between the heating area and the end area of ​​the insulator when there is a heating defect in the main body outline of the insulator, by using the cross-comparison ratio algorithm.

[0016] The heat defect type identification module is used to determine the heat defect type based on location-related parameters and preset heat defect classification configuration. The heat defect types include end heat and non-end heat.

[0017] The intersection-to-intersection ratio algorithm is configured to calculate the ratio of the intersection area of ​​the heating region and the insulator end region to the area of ​​the heating region.

[0018] According to a third aspect, the present invention provides an electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform the thermal defect type identification method for any composite insulator in the embodiments of the present invention.

[0019] According to another aspect of the present invention, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause a computer to execute a method for identifying the type of thermal defect in any composite insulator in the embodiments of the present invention.

[0020] The present invention provides a method for identifying the type of thermal defects in composite insulators. This method is achieved through five core steps: infrared image data acquisition, insulator body contour segmentation, thermal region extraction and defect judgment, spatial positional relationship quantitative analysis, and thermal defect type classification. Specifically, acquiring infrared image data of the target composite insulator is used to obtain raw detection data; performing insulator target detection and semantic segmentation based on the infrared image data to obtain the insulator body contour is used for precise target positioning and contour extraction; extracting the thermal region from the insulator body contour and determining whether a thermal defect exists based on temperature characteristics and a preset threshold is used for accurate identification of the thermal region; when a thermal defect exists, calculating position-related parameters based on the spatial positional relationship between the thermal region and the insulator end region using an intersection-union-ratio (IUU) algorithm is used for quantitative characterization of the thermal point's positional attributes; and determining the thermal defect type based on the position-related parameters and a preset thermal defect classification configuration is used for refined defect classification.

[0021] In this technical solution, the present invention addresses the problem described in the background art of coarse-grained defect identification and inability to distinguish defect types, leading to inaccurate risk assessment. It introduces an intersection-union-ratio (IUU) algorithm based on spatial location relationships to calculate location-related parameters, and uses these parameters in conjunction with a preset classification configuration to determine the type. This achieves automatic and accurate differentiation between "end-heating" and "non-end-heating" defects of different risk levels, solving the technical deficiency of existing technologies that can only determine "whether it is heating" but cannot perform risk tracing and classification. Addressing the problem described in the background art of lagging data processing and inability to support real-time on-site decision-making, the present invention constructs an integrated online analysis capability through a coherent and automated processing flow from acquiring infrared image data to outputting defect types. This provides a core processing foundation for subsequent real-time judgment and early warning, alleviating the technical drawbacks of the existing "collect first, process later" mode, such as large analysis delays and lack of on-site interaction. Therefore, the technical solution of the present invention solves the problem of insufficient accuracy in identifying the heating defect type of composite insulators in UAV inspection scenarios, improving the accuracy of heating defect identification and the reliability of risk classification. Attached Figure Description

[0022] Figure 1 This is a flowchart of a method for identifying the type of thermal defects in composite insulators according to an embodiment of the present invention; Figure 2 This diagram illustrates a specific system architecture and data processing flow according to another embodiment of the present invention. Figure 3 This diagram illustrates the hardware deployment and real-time data interaction architecture of a drone inspection system according to another embodiment of the present invention. Figure 4 This diagram illustrates the hardware architecture of an edge device used to perform the task of identifying the type of thermal defects in composite insulators, according to another embodiment of the present invention. Figure 5 This is a schematic diagram of the structure of a composite insulator thermal defect type identification system according to an embodiment of the present invention; Figure 6 This is a block diagram of an electronic device used to implement embodiments of the present invention. Detailed Implementation

[0023] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0024] During the development of this invention, the inventors, through extensive experiments and data analysis, revealed the intrinsic correlation between the type of thermal defects in composite insulators and the spatial location of the thermal region: non-end heating often indicates internal core rod defects and carries extremely high risk, while end heating is mostly caused by moisture in the sheath and carries lower risk. This difference in risk can be quantified by the degree of geometric overlap between the thermal region and the end region of the insulator. Based on this relationship, the inventors innovatively proposed this technical solution, utilizing infrared image data acquisition and edge intelligent analysis technology. Through target detection, semantic segmentation, thermal region extraction, and cross-parameter ratio (CPAR) algorithm calculation of position parameters, real-time and accurate identification of end heating and non-end heating is achieved, embodying the core concept of intelligent defect classification based on spatial location relationships.

[0025] Specifically, through comparative experiments, the invention team discovered that traditional composite insulator heating defect detection methods (such as those based on overall temperature threshold judgment) suffer from coarse-grained defect identification: they can only determine "whether it is heating up," but cannot distinguish whether the heating point is located at the end or not. These technical defects lead to inaccurate risk assessment, and maintenance personnel cannot quickly formulate differentiated maintenance strategies based on the detection results, seriously affecting the efficiency of power grid safe operation and the timeliness of fault early warning. However, the spatial location quantification analysis proposed in this invention (such as the cross-parallel comparison algorithm) can improve the accuracy and reliability of defect type identification; real-time acquisition and edge processing of infrared image data can enable immediate decision-making and interaction at the inspection site; and by calculating the position parameters of the heating area and the end area, high-risk non-end heating defects can be promptly warned and accurately controlled.

[0026] Therefore, this invention provides a method for identifying the type of thermal defects in composite insulators based on the first aspect. This method can be applied to an intelligent analysis system for UAV inspection of transmission lines (hereinafter referred to as the "system"). The system can run on an edge computing platform via embedded independent operation or cloud-based collaborative processing to complete real-time acquisition, analysis, and defect type identification of infrared images at the inspection site. Specifically, this system can be deployed in various hardware environments, including but not limited to: embedded edge intelligent computing devices mounted on UAV platforms, ground inspection mobile workstations, and edge servers in regional monitoring centers. This flexible deployment architecture allows the system to meet the stringent requirements of low latency and real-time processing at the inspection site, while also adapting to the data aggregation and batch analysis needs of different levels of operation and maintenance centers.

[0027] like Figure 1 As shown; the method may include: Step S110: Acquire infrared image data of the target composite insulator.

[0028] Infrared image data refers to digital image data reflecting the surface temperature distribution of composite insulators, acquired by an infrared thermal imager mounted on a drone. The data format is typically JPEG or TIFF, containing pixel-level temperature information. JPEG (Joint Photographic Experts Group) is a lossy compression digital image format standard widely used in photography and network transmission. TIFF (Tagged Image File Format) is a flexible lossless or lightly compressed image file format commonly used in professional image processing, printing, and archival storage.

[0029] Specifically, the system can use the image acquisition module deployed on the inspection platform to poll the drone's storage medium (such as an SD card, short for Secure Digital Card) in real time to acquire newly generated infrared image files. For example, the system polls the SD card file system every 2 seconds, identifies new files through difference operations, and downloads them to the edge computing device cache; the file size is approximately 2MB (megabytes).

[0030] Step S120: Perform insulator target detection and semantic segmentation operations based on infrared image data to obtain the main outline of the insulator.

[0031] Among them, the insulator target detection and semantic segmentation operation refers to using a deep learning model to automatically identify composite insulator targets in infrared images and perform pixel-level segmentation to extract accurate geometric contours; the main contour of the insulator refers to the binary mask or set of edge pixels of the composite insulator obtained after segmentation, which is used to characterize the complete shape of the insulator.

[0032] Specifically, the system can identify composite insulator targets using a pre-trained target detection neural network (such as YOLO, short for You Only Look Once), outputting an initial bounding box. Then, a semantic segmentation algorithm (such as a deep learning-based semantic segmentation model) is applied to perform pixel-level segmentation of the region within the bounding box, generating a binary mask and calculating the minimum bounding rectangle and its tilt angle. For example, the system uses the YOLOv8 model (a deep learning-based one-stage real-time target detection algorithm, an important evolution of the YOLO series models) for detection, obtaining a set of edge pixels after segmentation, and calculating the tilt angle of the minimum bounding rectangle. for ,in and It is the component of the vertical vector of the rectangle, and the angle unit is degrees.

[0033] Step S130: Extract the heating area from the outline of the insulator body, and determine whether there is a heating defect based on the comparison between the temperature characteristics of the heating area and the preset threshold.

[0034] Among them, the heating area refers to the local area on the surface of the insulator that is highlighted due to abnormal temperature rise; the temperature characteristic refers to the difference between the highest temperature of the heating area and the average temperature of the area; the preset threshold refers to the temperature difference threshold value used to determine the heating defect, and its value is set based on historical experimental data.

[0035] Specifically, the system can identify candidate hot spots within the smallest bounding rectangle of the insulator's main body outline, extract temperature data from the candidate spots using image processing technology, and calculate the temperature difference. (in This is the highest temperature in the heating area. (The average temperature of the smallest bounding rectangular region) is compared with a preset threshold. For example, when the system... When the Kelvin (K) value is reached, a heating defect is determined, and the boundary coordinates of the heating area are recorded in pixel position format. .

[0036] Step S140: If there is a heating defect in the main body outline of the insulator, the position-related parameters are calculated by the cross-comparison algorithm based on the spatial positional relationship between the heating area and the end area of ​​the insulator.

[0037] Among them, the insulator end region refers to the overall region formed by the equalizing ring of the high-voltage end of the composite insulator and the lower part of the first shed connected to it; the spatial positional relationship refers to the geometric overlap relationship between the heating region and the end region in the image; the positional related parameters are numerical indicators that quantify the degree of overlap between the two.

[0038] Specifically, the system can perform fine-grained region segmentation on the insulator's main outline to separate the end regions, then calculate the intersection area of ​​the heating region and the end regions, as well as the area of ​​the heating region itself, and obtain position-related parameters through ratio calculations. For example, the system uses... (An open-source computer vision and machine learning software library that provides a rich set of image processing, video analysis, and pattern recognition algorithms) Calculate the number of intersection pixels of boundary pixel sets, with the area in pixel squares. The ratio of the intersection area to the area of ​​the heating region ranges from [value missing]. .

[0039] Step S150: Determine the type of heat defect based on the location-related parameters and the preset heat defect classification configuration. The heat defect types include end heat and non-end heat.

[0040] The preset thermal defect classification configuration refers to a set of classification rules based on location-related parameters and the number of thermal points, used to distinguish defects of different risk levels.

[0041] Specifically, the system can determine the heating area by counting the number of heating regions: if the number is greater than or equal to 2, it is directly classified as non-end heating; if the number is 1, the system compares the location-related parameters with a preset threshold. If the parameters are greater than or equal to the threshold, it is end heating; otherwise, it is non-end heating (further verification is needed to confirm whether the highest temperature point is located within the end region). For example, the system sets the threshold to 0.1. When the location-related parameters are less than 0.1 and the highest temperature point of the heating region is not located within the end region, it is classified as non-end heating.

[0042] The intersection-to-intersection ratio algorithm is configured to calculate the ratio of the intersection area of ​​the heating region and the insulator end region to the area of ​​the heating region.

[0043] Specifically, the system can perform pixel-level geometric operations to first calculate the intersection of the pixel sets at the boundaries of the two regions, then calculate the number of pixels in the intersection and the number of pixels in the heated region, and finally the ratio is the quotient of the number of pixels in the intersection and the number of pixels in the heated region. For example, at an image resolution of 1024×768 pixels, if the area of ​​the heated region is 500 pixels square and the area of ​​the intersection is 50 pixels square, then the ratio is 0.1.

[0044] In other embodiments, such as Figure 2 The diagram illustrates a specific system architecture and data processing flow of the composite insulator heating defect type identification method of the present invention. The diagram clearly demonstrates the end-to-end processing logic and modular collaborative relationship from the input of raw infrared image data to the final visual output of the defect type.

[0045] Specifically, the system architecture and process begin with the real-time infrared image polling module. Deployed in an edge computing device mounted on the drone, this module proactively discovers and acquires newly captured infrared image files stored on the onboard SD card by the drone's infrared thermal imager through periodic file system scans (i.e., polling operations). For example, the module scans a specified directory every 2 seconds, identifies new files such as "DJI_20231027_143012.jpg", loads them into a memory buffer, and then processes them for subsequent modules. After acquiring the image, the process proceeds to the composite insulator detection and segmentation module. This module first uses a pre-trained target detection neural network (e.g., YOLOv8) to quickly infer the input infrared image, locates all composite insulator targets in the image, and outputs their initial bounding boxes. Subsequently, based on these bounding boxes, the region of interest is cropped, and a semantic segmentation algorithm (e.g., a model based on the DeepLab architecture) is used to perform pixel-level classification of this region, generating a binary mask that accurately outlines the insulator contours. For example, given a 1024×768 pixel image, this module outputs the bounding box coordinates of one or more insulators (e.g., ...). The module then analyzes the segmented insulator body contour and its corresponding binary mask matrix. Next, the heating defect judgment module analyzes the segmented insulator body contour. This module searches for clusters of bright pixels within the smallest bounding rectangle of the insulator as candidate heating points, extracts their temperature data, and calculates the difference between the highest temperature and the average temperature in that region. The presence of a heating defect is determined by comparing the result with a preset threshold (e.g., 5 Kelvin). If the threshold is exceeded, the boundary coordinates of the heated area are recorded. For example, if a heated area is detected, its... The value is 8.2 Kelvin, and the boundary coordinates are... If overheating is detected, the process is then handled by the composite insulator overheating type classification module for fine-grained identification. This module first performs secondary segmentation on the insulator outline, separating the end region (i.e., the equalizing ring and the lower part of the connected first shed). Then, it calculates the intersection ratio between the overheating region and this end region (…). ), as location-related parameters. Finally, based on preset rules (such as the number of heat points, etc. The value is compared with a threshold to determine whether the defect type is "end heating" or "non-end heating". For example, the calculated value is... The value is 0.05 (less than the threshold of 0.1), and the highest temperature point of the heating area is not located within the end region, therefore it is determined to be "non-end heating". After the determination is completed, the heating defect result push module encapsulates the identification results, including device ID, timestamp, insulator ID, defect type, and coordinates of the heating area, into structured data (such as JSON format, short for JavaScript Object Notation). The module pushes the data to the ground control terminal (such as a drone control handle) in real time through a pre-established dedicated wireless data channel based on the PSDK or MSDK protocol (Payload Software Development Kit, or Mobile Software Development Kit). For example, a JSON data packet is generated and sent through channel ID 0x01, with a transmission latency of less than 100 milliseconds. Finally, the defect receiving and display module of the ground terminal receives and parses the data packet. On the graphical user interface (GUI) of the terminal application, the identification results are displayed in real time to the drone pilot or maintenance personnel in a visual form (such as overlaying a highlighted box and type label on the original infrared image), thus completing the closed loop from image acquisition to on-site decision support. For example, in the infrared image on the screen, a red rectangle marks the location of the heat source, and the text "Alarm: Non-endpoint heat source" is displayed next to it.

[0046] Therefore, according to this embodiment, through the series connection and coordination of the above six core modules, this solution realizes online, automatic and accurate analysis of infrared images of UAV inspection, which can effectively distinguish the heating defects of different risk levels and provide key information support for on-site real-time decision-making.

[0047] In other embodiments, such as Figure 3 The diagram illustrates the hardware deployment and real-time data interaction architecture of the UAV inspection system involved in this invention. It clearly outlines the complete physical link and component collaboration relationships from infrared image acquisition and edge-side intelligent analysis to ground-side result reception.

[0048] Specifically, this architecture uses a drone as a carrier, with its onboard infrared thermal imager responsible for capturing infrared images of composite insulators on transmission lines. The raw image data is first temporarily stored on an onboard SD card. For example, the drone hovers near a transmission tower during its inspection route, photographs its B-phase composite insulator, generates an infrared image file named "DJI_0032.jpg", and saves it to the SD card. Subsequently, the data flows into the core processing unit—the onboard edge intelligent computing box. This box reads new image files from the SD card in real time via a high-speed data interface (such as USB 3.0) and carries all the algorithm modules from the aforementioned method embodiments, performing the task of determining the heating type of the infrared image. For example, the processor within the computing box calls a deep learning model to complete the insulator location, segmentation, heating type determination, and classification of "DJI_0032.jpg" within 2 seconds, arriving at a determination result of "non-end heating". After the determination result is generated, it is encapsulated and forwarded through the PSDK interface. The computing power box, acting as the PSDK payload device, establishes communication with the UAV flight control system and sends structured result data to the UAV data link via a designated channel ID (e.g., 0x01). This process establishes a dedicated uplink data channel from the payload to the flight control system. Data is then transmitted downlink to the UAV controller on the ground via the UAV's wireless link. The inspection application running on the controller listens for and receives data from the designated channel ID through the MSDK interface. For example, the controller application parses the received JSON data packets in real time, extracting information such as defect type and location. Finally, the results are presented to the pilot in real-time on the application interface (such as a map or real-time video overlay interface) in a visual format (e.g., pop-up alerts, icon markings). For example, the pilot sees an insulator highlighted in red with a message "Caution: Non-end heating (high risk)" on the controller screen, allowing immediate action for careful inspection or recording.

[0049] Therefore, according to this embodiment, through the collaboration of drones, airborne edge computing boxes, PSDK or MSDK communication protocols and ground control handles, this solution constructs a closed-loop system integrating image acquisition, real-time intelligent analysis and on-site decision feedback, realizing a mode innovation of power transmission line inspection from "traditional photo analysis" to "judgment while flying".

[0050] In other embodiments, such as Figure 4 The diagram illustrates the hardware architecture of the edge device used in this invention to perform the task of identifying the type of thermal defects in composite insulators. This diagram clearly shows the internal core components of the dedicated computing device designed to meet the real-time, intelligent analysis requirements of UAV inspection scenarios and their data interaction relationships.

[0051] Specifically, this edge device, serving as the core processing unit deployed on a drone platform, is built around a high-speed communication bus to enable low-latency data exchange between components. This communication bus is a bidirectional parallel or serial path for transmitting data, addresses, and control signals within a chip or between board-level components. For example, the bus can employ the AXI (Advanced eXtensible Interface) protocol, providing data transfer bandwidth of up to 32 GB / s (gigabytes per second) to ensure smooth data flow between processing units. Above the communication bus, an ARM architecture processor acts as the main control unit, responsible for running the device's operating system, task scheduling, and calling algorithm libraries to perform general computing tasks such as infrared image preprocessing, post-processing, and logic control. For example, an eight-core ARM Cortex-A78 processor with a clock speed of 2.0 GHz can efficiently coordinate the work of various modules and complete the format conversion and result encapsulation of an infrared image within 300 milliseconds (ms). Alongside the ARM processor, a memory unit stores system programs, deep learning models, image data to be processed, and intermediate calculation results. This memory typically includes RAM (such as LPDDR5, low-power double data rate fifth-generation synchronous dynamic random access memory) and persistent storage (such as eMMC, embedded multimedia card). For example, a device configured with 8GB of LPDDR5 RAM and 128GB of eMMC flash memory is sufficient to load approximately 50MB of YOLOv8 and DeepLab model parameters and batch image data. Below the bus, a dedicated NPU (Neural Processing Unit) is responsible for performing forward inference of deep learning models such as object detection and semantic segmentation, which is the most computationally intensive part of the entire recognition process. For example, an NPU with a computing power of 16 TOPS (trillion operations per second) can complete the insulator detection and segmentation task of a 1024×768 pixel infrared image in 150 milliseconds, achieving tens of times the energy efficiency improvement compared to a general-purpose processor. Also located below the bus, the communication interface unit is responsible for enabling data communication between the edge device and other subsystems of the UAV (such as flight control and gimbal camera) and storage media (such as SD card). This interface may include physical interfaces such as USB (Universal Serial Bus) and SDIO (Secure Digital Input and Output) and their controllers. For example, infrared images can be read from a drone's SD card via a USB 3.0 interface, and the recognition results can be written back to the memory card via the SDIO interface.

[0052] Therefore, according to this embodiment, by efficiently integrating an ARM architecture processor, NPU, memory, and communication interface through a communication bus, the edge device upon which this solution relies constructs a heterogeneous computing platform. This platform fully leverages the respective advantages of the ARM processor in process control and the NPU in AI inference, achieving real-time, high-precision intelligent analysis of infrared images of composite insulators under limited onboard space and power consumption constraints. This is the key hardware foundation for the successful deployment of this method at the inspection site.

[0053] Therefore, according to the above implementation method, the system achieves its purpose through five core steps: infrared image data acquisition, insulator body contour segmentation, heating area extraction and defect judgment, spatial position relationship quantitative analysis, and heating defect type classification. Specifically, acquiring infrared image data of the target composite insulator is used to obtain raw detection data; performing insulator target detection and semantic segmentation based on the infrared image data to obtain the insulator body contour is used to achieve precise target positioning and contour extraction; extracting the heating area from the insulator body contour and judging whether a heating defect exists based on temperature characteristics and a preset threshold is used to accurately identify the heating area; when a heating defect exists, calculating position-related parameters based on the spatial position relationship between the heating area and the insulator end area using the intersection-union-ratio algorithm is used to quantitatively characterize the heating point's position attributes; and determining the heating defect type based on the position-related parameters and a preset heating defect classification configuration is used to achieve refined defect classification.

[0054] Specifically, in this embodiment, the technical solution addresses the problem described in the background art: the existing technology's coarse-grained defect identification and inability to distinguish defect types, leading to inaccurate risk assessment. By introducing an intersection-union-ratio (IUU) algorithm based on spatial location relationships to calculate location-related parameters, and using these parameters in conjunction with a preset classification configuration for type determination, it achieves automatic and accurate differentiation between "end-heating" and "non-end-heating" defects of different risk levels. This solves the technical deficiency of existing technologies that can only determine "whether it is heating" but cannot perform risk tracing and classification. Addressing the problem described in the background art: the data processing flow is lagging and cannot support real-time on-site decision-making. Through a coherent and automated processing flow from acquiring infrared image data to outputting defect types, an integrated online analysis capability is constructed. This provides a core processing foundation for subsequent real-time judgment and early warning, alleviating the technical drawbacks of the existing "collect first, process later" mode, such as large analysis delays and lack of on-site interaction. Therefore, the technical solution of this embodiment solves the problem of insufficient accuracy in identifying the heating defect type of composite insulators in UAV inspection scenarios, improving the accuracy of heating defect identification and the reliability of risk classification.

[0055] In some embodiments, infrared image data is obtained by an image acquisition module deployed on an inspection platform through a polling operation; based on the infrared image data, insulator target detection and semantic segmentation operations are performed to obtain the main outline of the insulator, including: The target detection neural network identifies composite insulator targets in infrared images and outputs initial bounding boxes.

[0056] Here, the target detection neural network refers to a computer vision model built on a deep learning framework for automatically identifying specific targets in an image, such as the YOLO (You Only Look Once) series of networks; the initial bounding box refers to the rectangular region output by the model for coarsely locating the composite insulator target, and its coordinate format is... ,in and This represents the x and y coordinates of the top-left corner of the bounding box. and This represents the x and y coordinates of the bottom right corner, in pixels. Specifically, the system can load a pre-trained target detection model (such as YOLOv8) to perform forward inference on the input infrared image, extract multi-scale features, and output the bounding box coordinates and confidence score of the composite insulator through the detection head. For example, using the YOLOv8 model, the system can identify composite insulator targets in an infrared image with a resolution of 1024×768 pixels, and output the initial bounding box coordinates as follows: The confidence level is 0.95.

[0057] The semantic segmentation algorithm is used to segment the region within the initial bounding box at the pixel level, generating a binary mask for the composite insulator. The semantic segmentation algorithm is configured to classify the pixels within the region of the initial bounding box into either the composite insulator region or the background region, and generate the corresponding binary mask.

[0058] Pixel-level segmentation refers to the operation of classifying each pixel in an image; a binary mask is a matrix containing only 0 and 1 values, where 1 indicates that the pixel belongs to the composite insulator region and 0 indicates that the pixel belongs to the background region.

[0059] Specifically, the system can use a semantic segmentation model (such as a model based on U-Net or DeepLab architecture) to perform fine segmentation of the region within the initial bounding box, outputting a binary mask with the same resolution as the input region. For example, the system segments the initial bounding box region (200×150 pixels) mentioned above and generates a binary mask matrix. ,in Represents pixels It belongs to the composite insulator region.

[0060] Furthermore, U-Net is a convolutional neural network with an encoder-decoder structure. Its name derives from its unique U-shaped symmetric structure, initially widely used in biomedical image segmentation. This architecture captures contextual information of an image through an encoder (downsampling path), then achieves precise pixel localization through a decoder (upsampling path), and combines skip connections to fuse the high-resolution features of the encoder with the corresponding layers of the decoder, thereby achieving fine edge segmentation while reducing information loss. DeepLab is a semantic segmentation architecture based on deep convolutional neural networks and incorporating multi-scale contextual information. Its core innovation lies in the introduction of dilated convolution (or dilated convolution) to expand the receptive field without increasing the number of parameters or losing resolution, and the Atrous Spatial Pyramid Pooling (ASPP, a technique used in deep learning models for semantic segmentation) module to capture multi-scale contextual information in parallel. Subsequent versions (such as DeepLabv3+) also introduced an encoder-decoder structure to optimize the segmentation effect of object boundaries.

[0061] Extracting the set of edge pixels of composite insulators based on binary masks.

[0062] Among them, the edge pixel set refers to the sequence of pixel coordinates on the outline of the composite insulator region in the binary mask, which is used to characterize the geometric shape of the insulator.

[0063] Specifically, the system can traverse the binary mask using image processing algorithms (such as Canny edge detection or contour extraction algorithms, a multi-stage algorithm for extracting high-quality edges from digital images) to identify the boundary pixels of the composite insulator region and record their coordinates. For example, the system uses the findContours function from the OpenCV library (a computer vision algorithm for extracting object contours from binary images) to extract the set of edge pixels from the binary mask. ,in For example, the number of edge points. There are 10 points, and the coordinates of each point are in pixels.

[0064] Calculate the minimum bounding rectangle of the edge pixel set and obtain the tilt angle of the minimum bounding rectangle.

[0065] The minimum bounding rectangle is the smallest rectangle that can completely enclose the set of edge pixels; the tilt angle is the angle between the longer side of this rectangle and the vertical axis (y-axis) of the image, measured in degrees. ).

[0066] Specifically, the system can calculate the minimum bounding rectangle of the edge point set using geometric algorithms (such as the rotating caliper method), and calculate the tilt angle based on the coordinates of the rectangle's vertices. The formula is ,in and These are the components of the vertical vector of the rectangle. For example, the system calculates the coordinates of the four vertices of the smallest bounding rectangle as follows: tilt angle Spend.

[0067] Therefore, according to the above implementation method, the system can efficiently and accurately locate and extract the contour of composite insulators, providing a reliable geometric basis for subsequent analysis of heating defects.

[0068] In some embodiments, the heating region is extracted from the outline of the insulator body, and the presence of a heating defect is determined by comparing the temperature characteristics of the heating region with a preset threshold, including: Candidate hot spots are identified within the smallest circumscribed rectangular region of the insulator body profile.

[0069] Among them, candidate hot spots refer to local areas on the surface of composite insulators that appear brightly displayed in infrared images due to abnormal heating, with their temperature values ​​significantly higher than the surrounding background.

[0070] Specifically, the system can scan the minimum bounding rectangle region using a pre-trained convolutional neural network model to identify clusters of pixels with temperatures higher than the surrounding environment as candidate heat points. For example, using the YOLOv8 model for object detection, the system identifies a candidate heat point within the minimum bounding rectangle region, with its center pixel coordinates as follows: The coordinate unit is pixels.

[0071] Temperature data of candidate heating points are extracted, and the difference between the highest temperature of the candidate heating point and the average temperature of the smallest bounding rectangle region is calculated using a temperature difference calculation algorithm. The temperature difference calculation algorithm is configured to calculate the difference between the highest temperature of the candidate heating point and the average temperature of the smallest bounding rectangle region.

[0072] Among them, the temperature difference calculation algorithm is a mathematical operation used to quantify the degree of local heating. Its input is the temperature data of the candidate heating point and the average temperature of the smallest bounding rectangle area, and the output is the temperature difference value.

[0073] Specifically, the system obtains the temperature value of each pixel by calling the SDK (Software Development Kit) of the drone's infrared camera, and then calculates the highest temperature of all pixels within the candidate hot spot area. Simultaneously calculate the average temperature of all pixels within the smallest bounding rectangle region. Finally, the difference between the highest temperature and the average temperature is calculated. For example, for a candidate heat source, the highest temperature... The average temperature is 45.6 degrees Celsius. If the temperature is 37.4 degrees Celsius, then the difference is 8.2 Kelvin (K).

[0074] If the difference exceeds a preset threshold, a heating defect is determined, and the boundary coordinates of the heating area are recorded.

[0075] Among them, the preset threshold refers to the temperature difference threshold value set in advance for judging the heating defect, and its value is determined based on historical experimental data; the boundary coordinates refer to the range of pixel positions of the heating area in the image, usually represented by the coordinates of the upper left and lower right corners of the rectangle.

[0076] Specifically, the system compares the calculated difference with a preset threshold. If the difference is greater than or equal to the preset threshold (e.g., 5 Kelvin), the candidate heating point is determined to be a real heating defect, and the boundary coordinates of the heating area are recorded. For example, if the preset threshold is 5 Kelvin, and the difference is 8.2 Kelvin, the system determines that a heating defect exists and records the boundary coordinates of the heating area. , where 100 and 50 represent the x and y coordinates of the top left corner of the rectangle, and 200 and 150 represent the x and y coordinates of the bottom right corner, with the coordinate unit being pixels.

[0077] Therefore, according to the above implementation method, the system can automatically and accurately identify the thermal defects of composite insulators and provide reliable temperature characteristics and location data for subsequent defect type classification.

[0078] In some embodiments, the step of calculating position-related parameters using an intersection-exchange ratio algorithm based on the spatial relationship between the heating region and the insulator end region includes: The main outline of the insulator is divided into regions to separate the end regions of the insulator.

[0079] The insulator end region refers to the specific geometric region formed by the equalizing ring at the high-voltage end of the composite insulator and the lower part of the first shed connected to it. This region has unique shape and temperature characteristics in infrared images, which are used to distinguish end heating defects from non-end heating defects.

[0080] Specifically, the system can perform fine-grained secondary segmentation of the insulator's main outline using a pre-trained semantic segmentation model (such as a neural network based on the DeepLab architecture), classifying pixels into end regions or non-end regions and generating corresponding binary masks. For example, using the DeepLabv3+ model, the system takes the minimum bounding rectangle region of the insulator's main outline (200 pixels × 150 pixels) as input and outputs a binary mask of the end regions, where the pixel value of the end regions is 1 and the pixel value of the non-end regions is 0.

[0081] The intersection area between the heating region and the insulator end region is calculated based on the intersection operation of the boundary pixel set of the heating region and the boundary pixel set of the insulator end region.

[0082] Among them, the boundary pixel set refers to the sequence of continuous pixel coordinates that represent the contour of a region, extracted by an image processing algorithm; the intersection operation is a mathematical operation that calculates the overlapping part of two geometric regions and is used to quantify the degree of spatial overlap.

[0083] Specifically, the system can use contour processing functions from computer vision libraries (such as OpenCV) to extract the boundary pixel sets of the heating region and the insulator end region, respectively. Then, it calculates the number of overlapping pixels in the two sets; the intersection area is the total number of pixels in the intersection, expressed in pixels squared. For example, if the boundary pixel set of the heating region contains 500 points and the boundary pixel set of the insulator end region contains 300 points, the intersection operation will result in 80 overlapping pixels, and the intersection area will be 80 pixels squared.

[0084] The area of ​​the heated region is calculated based on the area of ​​the boundary pixel set of the heated region.

[0085] Area calculation refers to the operation of quantifying the size of a region by counting the number of pixels within the region.

[0086] Specifically, the system can generate a region-filling mask from the boundary pixel set and count the number of pixels with a value of 1 in the mask; this number represents the area of ​​the heat-generating region. For example, if the boundary pixel set of the heat-generating region is filled to generate a mask, and the count is 1200 pixels, then the area of ​​the heat-generating region is 1200 pixels squared.

[0087] The location-related parameters are obtained by calculating the ratio of the intersection area to the area of ​​the heating region.

[0088] Among them, the ratio operation refers to the calculation process of dividing the intersection area by the area of ​​the heating region, which is used to generate a dimensionless similarity metric value, which ranges from 0 to 1.

[0089] Specifically, the system can calculate the ratio using arithmetic division, with the following formula: For example, if the intersection area is 80 pixels squared and the heating area is 1200 pixels squared, then the location-related parameters are: .

[0090] Therefore, according to the above implementation method, the system can accurately quantify the degree of spatial overlap between the heating area and the insulator end area, providing a reliable geometric parameter basis for defect type classification.

[0091] In some embodiments, the type of thermal defect is determined based on location-related parameters and a preset thermal defect classification configuration, including: Determine whether there are multiple fever areas.

[0092] The number of heating areas refers to the total number of independent heating points identified in the infrared image, with each heating point corresponding to a continuous local high-temperature area.

[0093] Specifically, the system can process the extracted binary mask of the hot region using a connected component analysis algorithm (such as an eight-neighborhood-based connected component labeling algorithm) and count the number of connected regions as the number of hot regions. For example, if the system detects that the binary mask of the hot region contains two independent connected regions, then the number of hot regions is 2.

[0094] If there are multiple heating areas, the heating defect type is determined to be non-end heating.

[0095] Non-end heating refers to a defect type where the heating point is located in the non-end region of the composite insulator (such as the middle of the core rod). Multiple heating points usually indicate distributed defects or internal core rod problems, and the risk level is high.

[0096] Specifically, the system directly labels the heating defect type as "non-end heating" based on the logical condition that the number of heating areas is greater than or equal to 2, without further calculation of location-related parameters. For example, when the number of heating areas is 3, the system directly outputs the defect type as "non-end heating".

[0097] If the number of heating areas is single, the location-related parameters are compared with a preset threshold. If the location-related parameters are greater than or equal to the preset threshold, the heating defect type is determined to be end heating. If the location-related parameters are less than the preset threshold, and the highest temperature point of the heating area is not located in the end area of ​​the insulator, the heating defect type is determined to be non-end heating.

[0098] Among them, the position-related parameter refers to the value calculated by the cross-union ratio algorithm to quantify the degree of spatial overlap between the heating area and the end area of ​​the insulator; the preset threshold refers to the classification threshold value set based on historical experimental data, used to distinguish between end and non-end heating; the highest temperature point refers to the coordinates of the pixel point with the highest temperature value in the heating area; the end area of ​​the insulator refers to the geometric area formed by the equalizing ring of the high voltage end of the composite insulator and the lower part of the first shed connected to it.

[0099] Specifically, the system first checks whether location-related parameters (such as the cross-parallel ratio) are greater than or equal to a preset threshold (e.g., 0.1). If the condition is met, it is determined to be end-heating. If not, it further checks whether the coordinates of the highest temperature point are within the boundary range of the insulator end region (determined by coordinate comparison). If it is not within the end region, it is determined to be non-end-heating. For example, the location-related parameter is 0.05 (less than the threshold 0.1), and the coordinates of the highest temperature point... If the defect is not within the boundary of the end region, the system determines the defect type as "non-end heating".

[0100] Therefore, according to the above implementation method, the system can automatically and accurately distinguish the types of thermal defects in composite insulators, thereby improving the accuracy of risk assessment.

[0101] In some embodiments, after determining the type of thermal defect based on location-related parameters and a preset thermal defect classification configuration, the method further includes: The identification results of the heat-generating defect type are encapsulated into a structured data format, which includes the device identifier, timestamp, insulator identifier, defect type, and heat-generating area parameters.

[0102] Among them, structured data format refers to a standard data organization method used for machine-readable and data exchange, such as JSON format; device identifier refers to a string code that uniquely identifies a drone or edge device; timestamp refers to the precise time record when the data was generated, using the ISO 8601 standard format; insulator identifier refers to the unique identification code of a composite insulator; defect type refers to the classification result of the identified heating defects; heating area parameters refer to the set of values ​​that describe the geometric location and temperature characteristics of the heating area.

[0103] Specifically, the system can convert the recognition results into a JSON object using a data serialization module. This object contains a key-value pair structure, where the keys are predefined fields and the values ​​are the corresponding data. For example, the system generates a JSON object: {"device_id":"UAV_001","timestamp":"2023-10-27T14:30:15.123Z","insulator_id":"TL-110kV-0056-A phase","defect_type":"non-end heating","hot_spots":[{"spot_id":1,"bounding_box":[100,50,200,150],"max_temp":45.6,"avg_delta_t":8.2}]}, where bounding_box coordinates are in pixels and temperature is in degrees Celsius. ).

[0104] The encapsulated structured data is pushed to the ground control terminal via a pre-set communication link.

[0105] The preset communication link refers to a dedicated wireless channel established based on the DJI PSDK (Payload Software Development Kit) and MSDK (Mobile Software Development Kit) protocols for data transmission between the drone and ground equipment; the ground control terminal refers to devices such as drone handles or ground workstations that have data receiving and display capabilities.

[0106] Specifically, the system can use a dedicated data channel established between the edge computing device and the drone control handle to monitor newly generated structured data files using a polling mechanism, and then transmit them point-to-point through a specified channel ID. For example, the system polls the local storage of the edge device every 2 seconds, and after detecting a new JSON file, it pushes it to the ground control terminal through channel ID 0x01, with a transmission latency of less than 100 milliseconds (ms).

[0107] The structured data is parsed at the ground control terminal and the recognition results are displayed in a visual form.

[0108] Among them, visualization refers to presenting data in an intuitive graphic and textual way through a graphical user interface; parsing refers to the operation of decoding and extracting key information from the received structured data.

[0109] Specifically, the system can use an inspection application running on the ground control terminal to call a JSON parsing library (such as Java's Jackson library, used for serialization and deserialization between Java objects and JSON data formats) to parse data and overlay infrared images, heat-generating area markings, and defect type labels on the interface. For example, after parsing the JSON, the application marks the heat-generating area (coordinates) with a red rectangle on the infrared image. The text label "Non-end heating" is displayed next to the box, and the timestamp "2023-10-27T14:30:15.123Z" is recorded.

[0110] Therefore, according to the above implementation method, the system can realize automated encapsulation of recognition results, low-latency transmission and real-time visualization, thereby improving the efficiency of on-site decision-making during inspections.

[0111] In some embodiments, the encapsulated structured data is pushed to a ground control terminal via a preset communication link, including: Establish a dedicated data channel between edge computing devices and drone control handles.

[0112] Among them, the dedicated data channel refers to a point-to-point wireless communication link with a unique channel identifier established based on DJI's PSDK and MSDK protocols, used to realize low-latency data transmission between edge computing devices and drone controllers.

[0113] Specifically, the system can allocate a dedicated channel ID (identifier) ​​between the UAV flight control system and the edge computing device by calling the PSDK interface, and use this ID to initialize the communication session and establish a stable bidirectional data link. For example, the system allocates a channel ID of 0x01 and ensures link reliability through a physical connection of E-port coaxial cable, with a transmission latency of less than 100 milliseconds (ms).

[0114] The system monitors the files awaiting push notifications stored locally on edge computing devices using a polling mechanism.

[0115] Among them, the polling mechanism refers to a computer program logic that periodically and actively queries the file system for changes, used to detect newly generated files in real time; the file to be pushed is a structured data file that encapsulates the results of the thermal defect identification, and its format is JSON.

[0116] Specifically, the system can trigger a file system scan every 2 seconds via an embedded timer, traversing a specified directory (such as " / cache / results") in the local storage of the edge computing device. It compares the current file list with the list of processed file records and identifies new files through a difference operation. For example, after scanning the directory, the system obtains a file list Files_Current = {"result_001.json","result_002.json"} and a processed list Files_Processed = {"result_001.json"}. The difference operation then yields a new file Files_New = {"result_002.json"}.

[0117] In response to the detection of a newly generated structured data file, it is transmitted to the ground control terminal via a dedicated data channel.

[0118] Among them, the structured data file refers to the thermal defect identification result data encapsulated in JSON format, which includes fields such as device identifier, timestamp, and defect type; the ground control terminal refers to the inspection application interface running on the drone handle or ground workstation.

[0119] Specifically, the system can load the content of newly detected JSON files through the file reading interface, and encapsulate the data packets into a specified protocol format (such as a binary stream) through the PSDK data sending interface, and perform point-to-point transmission using the established channel ID. For example, if the system detects a new file "result_002.json" (approximately 2KB in size, kilobytes), it can transmit it to the drone controller through channel ID 0x01 at a transmission rate of up to 1Mbps (megabits per second).

[0120] Update the list of processed file records to synchronize data status.

[0121] The processed file record list refers to a dynamically maintained text or database record used to track the filenames and statuses of successfully transmitted files, ensuring data consistency.

[0122] Specifically, the system can append the newly successfully transmitted filename to the processed list through file system operations or database update commands, and record the timestamp and transmission status flag. For example, after successfully transmitting "result_002.json", the system updates the list to Files_Processed = {"result_001.json","result_002.json"}, and records the update timestamp "2023-10-27T14:30:15.123Z".

[0123] Therefore, according to the above implementation method, the system can realize the automation of identification results, low-latency transmission and status synchronization, and improve the real-time performance and reliability of data interaction at the inspection site.

[0124] In other embodiments, the system can complete the task of encapsulating and pushing the results of thermal defect identification through the following code example: { "device_id":"UAV_001", "timestamp":"2023-10-27T14:30:15.123Z", "insulator_id":"TL-110kV-0056-A phase", "defect_type":"Non-end heating", "hot_spots": [ { "spot_id": 1, "bounding_box": [x1, y1, x2, y2], "max_temp": 45.6, "avg_delta_t": 8.2 } ], "confidence": 0.96 } This code example is a structured data template based on JSON format, used to standardize and encapsulate the complete results of composite insulator heating defect identification; device_id refers to a string code that uniquely identifies the drone or edge device; timestamp refers to the precise time record when the data was generated, using the ISO 8601 standard format; insulator_id refers to the unique identification code of the composite insulator; defect_type refers to the identified heating defect classification result, such as "end heating" or "non-end heating"; hot_spots refers to a list structure used to store detailed information of one or more heating areas; spot_id refers to the unique identifier of the heating area; bounding_box refers to the pixel coordinate range of the heating area in the image, in the format [x1,y1,x2,y2], where x1 and y1 represent the horizontal and vertical coordinates of the upper left corner of the rectangle, and x2 and y2 represent the horizontal and vertical coordinates of the lower right corner, with the coordinate unit being pixels; max_temp refers to the highest temperature value of the heating area, in degrees Celsius. ); avg_delta_t refers to the difference between the highest temperature in the heating area and the average temperature of the area, in Kelvin (K); confidence refers to the model's confidence score for the recognition results, with a value range of... .

[0125] Specifically, the system can automatically generate the aforementioned JSON structure on the edge intelligent computing device through the heating defect result push module. The module first receives diagnostic result data from the composite insulator heating type classification module, including defect type, heating area coordinates, temperature characteristics, etc.; then it fills the JSON template according to predefined key-value pair rules; finally, it saves the encapsulated data as a local file and pushes it to the ground control terminal through the wireless communication link.

[0126] For example, after the system identifies a non-end heating defect in a composite insulator, the following JSON file is generated: device_id is filled with "UAV_001" to indicate the drone number; timestamp is filled with "2023-10-27T14:30:15.123Z" to indicate Coordinated Universal Time accurate to milliseconds; insulator_id is filled with "TL-110kV-0056-A phase" to indicate the transmission line tower number and phase; defect_type is filled with "non-end heating"; the hot_spots list contains one element, where spot_id is 1 and bounding_box is... This indicates the pixel range of the heating area; max_temp = 45.6 indicates a maximum temperature of 45.6 degrees Celsius; avg_delta_t = 8.2 indicates a temperature difference of 8.2 Kelvin; confidence = 0.96 indicates a confidence level of 96%. This JSON instance standardizes and machine-readables the recognition results, providing a unified format foundation for subsequent data transmission and visualization.

[0127] Therefore, according to the above implementation method, the system can automatically and efficiently complete the encapsulation and push of heat defect results, and improve the standardization and interoperability of inspection data processing.

[0128] Figure 5 This is a structural block diagram of a composite insulator thermal defect type identification system according to an embodiment of the present invention.

[0129] like Figure 5 As shown, the thermal defect type identification system for the composite insulator includes: The infrared image data acquisition module 210 is used to acquire infrared image data of the target composite insulator.

[0130] The insulator body contour segmentation module 220 is used to perform insulator target detection and semantic segmentation operations based on infrared image data to obtain the insulator body contour.

[0131] The heating defect judgment module 230 is used to extract the heating area from the outline of the insulator body and judge whether there is a heating defect based on the comparison between the temperature characteristics of the heating area and a preset threshold.

[0132] The position-related parameter calculation module 240 is used to calculate the position-related parameters based on the spatial positional relationship between the heating area and the end area of ​​the insulator when there is a heating defect in the main body outline of the insulator. The calculation is performed by the cross-combination ratio algorithm.

[0133] The heat defect type identification module 250 is used to determine the heat defect type based on location-related parameters and preset heat defect classification configuration. The heat defect types include end heat and non-end heat.

[0134] The intersection-to-intersection ratio algorithm is configured to calculate the ratio of the intersection area of ​​the heating region and the insulator end region to the area of ​​the heating region.

[0135] The specific functions and examples of each module and submodule of the device in this embodiment of the invention can be found in the relevant descriptions of the corresponding steps in the above method embodiments, and will not be repeated here.

[0136] According to embodiments of the present invention, the above-described method of the present invention can be applied to an electronic device and a readable storage medium.

[0137] Figure 6 A schematic block diagram of an example electronic device 600 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0138] like Figure 6 As shown, the electronic device 600 includes a computing unit 601, which can perform various appropriate actions and processes based on a computer program stored in a read-only memory (ROM) 602 or a computer program loaded from a storage unit 608 into a random access memory (RAM) 603. The RAM 603 may also store various programs and data required for the operation of the electronic device 600. The computing unit 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0139] Multiple components in electronic device 600 are connected to I / O interface 605, including: input unit 606, such as keyboard, mouse, etc.; output unit 607, such as various types of displays, speakers, etc.; storage unit 608, such as disk, optical disk, etc.; and communication unit 609, such as network card, modem, wireless transceiver, etc. Communication unit 609 allows electronic device 600 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0140] The computing unit 601 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 601 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 601 performs the various methods and processes described above, such as a method for identifying the type of thermal defects in composite insulators. For example, in some embodiments, a method for identifying the type of thermal defects in composite insulators can be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as storage unit 608. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 600 via ROM 602 and / or communication unit 609. When the computer program is loaded into RAM 603 and executed by the computing unit 601, one or more steps of the method for identifying the type of thermal defects in composite insulators described above can be performed. Alternatively, in other embodiments, the computing unit 601 may be configured by any other suitable means (e.g., by means of firmware) to perform a method for identifying the type of thermal defects in composite insulators.

[0141] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0142] The program code used to implement the methods of the present invention can be written in any combination of one or more programming languages. This program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely on the machine, partially on the machine, as a standalone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0143] In the context of this invention, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can include, but are not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0144] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0145] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0146] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0147] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0148] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method for identifying the type of thermal defect in a composite insulator, characterized in that, include: Acquire infrared image data of the target composite insulator; Based on the infrared image data, insulator target detection and semantic segmentation operations are performed to obtain the main outline of the insulator; The heating area is extracted from the outline of the insulator body, and the presence of heating defects is determined by comparing the temperature characteristics of the heating area with a preset threshold. If there is a heating defect in the outline of the insulator body, the position-related parameters are calculated by the cross-union ratio algorithm based on the spatial positional relationship between the heating area and the end area of ​​the insulator. The type of heat defect is determined based on the location-related parameters and the preset heat defect classification configuration. The heat defect types include end heat and non-end heat. The intersection-to-intersection ratio algorithm is configured to calculate the ratio of the intersection area of ​​the heating region and the end region of the insulator to the area of ​​the heating region.

2. The method according to claim 1, characterized in that, The infrared image data is obtained by the image acquisition module deployed on the inspection platform through a polling operation; the insulator target detection and semantic segmentation operation is performed based on the infrared image data to obtain the main outline of the insulator, including: The target detection neural network identifies composite insulator targets in the infrared image and outputs an initial bounding box. The region within the initial bounding box is segmented at the pixel level using a semantic segmentation algorithm to generate a binary mask for the composite insulator. The semantic segmentation algorithm is configured to classify the pixels within the region of the initial bounding box into either a composite insulator region or a background region, and generate the corresponding binary mask. Extract the set of edge pixels of the composite insulator based on the binary mask; Calculate the minimum bounding rectangle of the edge pixel set, and obtain the tilt angle of the minimum bounding rectangle.

3. The method according to claim 1, characterized in that, The step of extracting the heating region from the outline of the insulator body and determining whether a heating defect exists based on a comparison of the temperature characteristics of the heating region with a preset threshold includes: Candidate hot spots are identified within the smallest circumscribed rectangular area of ​​the insulator body contour; The temperature data of the candidate heating points are extracted, and the difference between the highest temperature of the candidate heating points and the average temperature of the minimum bounding rectangle region is calculated using a temperature difference calculation algorithm. The temperature difference calculation algorithm is configured to calculate the difference between the highest temperature of the candidate heating points and the average temperature of the minimum bounding rectangle region. In response to the difference exceeding a preset threshold, a heating defect is determined to exist, and the boundary coordinates of the heating area are recorded.

4. The method according to claim 1, characterized in that, The step of calculating position-related parameters based on the spatial relationship between the heating region and the insulator end region using the cross-comparison ratio algorithm includes: The outline of the insulator body is divided into regions to separate the end regions of the insulator; The intersection area of ​​the heating region and the insulator end region is calculated based on the intersection operation of the boundary pixel set of the heating region and the boundary pixel set of the insulator end region. The area of ​​the heated region is calculated based on the area of ​​the boundary pixel set of the heated region. The location-related parameters are obtained by calculating the ratio of the intersection area to the area of ​​the heating region.

5. The method according to claim 1, characterized in that, The step of determining the type of heat defect based on the location-related parameters and the preset heat defect classification configuration includes: Determine whether there are multiple heating areas; If there are multiple heating areas, the heating defect type is determined to be non-end heating. If the number of heating areas is single, the location-related parameter is compared with a preset threshold; if the location-related parameter is greater than or equal to the preset threshold, the heating defect type is determined to be end heating; if the location-related parameter is less than the preset threshold, and the highest temperature point of the heating area is not located in the end area of ​​the insulator, the heating defect type is determined to be non-end heating.

6. The method according to claim 1, characterized in that, After determining the type of heat defect based on the location-related parameters and the preset heat defect classification configuration, the method further includes: The identification results of the heat-generating defect type are encapsulated into a structured data format, which includes device identifier, timestamp, insulator identifier, defect type, and heat-generating area parameters. The encapsulated structured data is pushed to the ground control terminal through a pre-set communication link; The ground control terminal parses the structured data and displays the recognition results in a visual form.

7. The method according to claim 6, characterized in that, The step of pushing the encapsulated structured data to the ground control terminal via a preset communication link includes: Establish a dedicated data channel between edge computing devices and drone control handles; The system monitors the files of results to be pushed that are stored locally on the edge computing device through a polling mechanism. In response to the detection of the newly generated structured data file, it is transmitted to the ground control terminal via the dedicated data channel; Update the list of processed file records to synchronize data status.

8. A system for identifying the type of thermal defects in composite insulators, characterized in that, include: Infrared image data acquisition module, used to acquire infrared image data of the target composite insulator; The insulator body contour segmentation module is used to perform insulator target detection and semantic segmentation operations based on the infrared image data to obtain the insulator body contour. The heating defect judgment module is used to extract the heating area from the outline of the insulator body and judge whether there is a heating defect based on the comparison between the temperature characteristics of the heating area and a preset threshold. The position-related parameter calculation module is used to calculate the position-related parameters based on the spatial positional relationship between the heating area and the end area of ​​the insulator when there is a heating defect in the outline of the main body of the insulator, by using the cross-union ratio algorithm. The heat defect type identification module is used to determine the heat defect type based on the location-related parameters and the preset heat defect classification configuration. The heat defect type includes end heat and non-end heat. The intersection-to-intersection ratio algorithm is configured to calculate the ratio of the intersection area of ​​the heating region and the end region of the insulator to the area of ​​the heating region.

9. An electronic device, characterized in that, include: At least one processor; and a memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, in, Computer instructions are used to cause a computer to perform the method according to any one of claims 1-7.

Images