Power equipment insulation defect image recognition system and method based on deep learning

Abstract

This invention discloses a deep learning-based image recognition system and method for insulation defects in power equipment, relating to the field of intelligent detection technology for power equipment. The system includes: receiving an original image and dividing it into foreground, midground, and background regions; extracting electrical topology data and generating a discharge path risk heatmap by combining it with the midground region; identifying candidate defect regions in the foreground and midground using a target detection model; extracting discharge risk distribution data of the defect regions by combining the heatmap; analyzing background image features to estimate image noise probability; fusing discharge risk and noise probability through a preset risk confidence calculation function; calculating a defect risk confidence score; and marking candidate defect regions in the image whose risk confidence scores are higher than a threshold. Its beneficial effects are: achieving spatial and quantitative risk judgment of defects through the discharge path risk heatmap, improving the accuracy and robustness of insulation defect identification, and enabling precise labeling of high-risk defects.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection technology for power equipment, and in particular to an image recognition system and method for insulation defects in power equipment based on deep learning. Background Technology

[0002] With the expansion of power system scale and the increasing demand for intelligent inspection, deep learning-based image recognition methods have become an important means of detecting surface defects in equipment such as insulators, bushings, and drain clamps. The typical process is: inspection terminal acquires high-definition images → convolutional neural network extracts surface texture features such as cracks and creepage marks → target detection or segmentation network marks the defect location. This technology significantly reduces the intensity of manual inspection, improves detection efficiency, and has been verified in a large number of transmission lines and substations.

[0003] Existing image-based insulation defect identification methods still have the following shortcomings in engineering applications: the detection results only reflect the appearance of the defect and cannot spatially and quantitatively assess the discharge risk of the defect. In actual power equipment, the discharge risk of the same type of surface defect located at different electrical nodes may differ by several times, and existing technologies are unable to distinguish high-risk defects, making it impossible for maintenance personnel to accurately formulate maintenance plans based on the identification results.

[0004] Therefore, a deep learning-based image recognition method for insulation defects in power equipment is proposed. Summary of the Invention

[0005] In view of the above-mentioned prior art, this application is hereby filed. Embodiments of this application provide a deep learning-based image recognition system and method for insulation defects in power equipment, which can improve the accuracy and robustness of insulation defect recognition and achieve precise labeling of high-risk defects.

[0006] According to one aspect of this application, a deep learning-based image recognition method for insulation defects in power equipment is provided, comprising: receiving an original image; dividing the original image into a foreground region, a midground region, and a background region, wherein the foreground region is the main body of the insulating equipment, the midground region is electrical components connected to the main body, and the background region is an image portion unrelated to the main body; extracting electrical topology data corresponding to the main body, the electrical topology data including electrical connection information and electrical parameters between the main body and the electrical components; and constructing a discharge path risk heatmap corresponding to the space of the original image based on the electrical topology data and the midground region, wherein the discharge path risk heatmap... The image includes the discharge probability distribution and spatial location between electrical nodes; target detection is performed on the foreground and midground regions using a deep learning-based target detection model to obtain candidate defect regions; discharge risk distribution data for each candidate defect region is extracted based on the discharge path risk heatmap; image feature information of the background region is extracted, and image noise probability of the candidate defect regions is calculated based on the image feature data; risk confidence score of the candidate defect regions is calculated based on a preset risk confidence calculation function, combined with the discharge risk distribution data and the image noise probability; candidate defect regions with risk confidence scores higher than a preset score threshold are marked in the original image.

[0007] According to another aspect of this application, a deep learning-based image recognition system for insulation defects in power equipment is provided, comprising: an image receiving module for receiving an original image; a region segmentation module for dividing the original image into a foreground region, a midground region, and a background region, wherein the foreground region is the main body of the insulation equipment, the midground region is an electrical component connected to the main body, and the background region is an image portion unrelated to the main body; an electrical topology extraction module for extracting electrical topology data corresponding to the main body, wherein the electrical topology data includes electrical connection relationship information and electrical parameters between the main body and the electrical component; and a risk heatmap construction module for constructing a discharge path risk heatmap corresponding to the space of the original image based on the electrical topology data and the midground region, wherein the discharge path risk heatmap includes electrical... The system includes: a discharge probability distribution and spatial location between gas nodes; a target detection module for detecting targets in the foreground and midground regions using a deep learning-based target detection model to obtain candidate defect regions; a risk mapping module for extracting discharge risk distribution data for each candidate defect region based on the discharge path risk heatmap; an image feature extraction module for extracting image feature information from the background region and calculating the image noise probability of the candidate defect regions based on the image feature data; a risk confidence calculation module for calculating the risk confidence score of the candidate defect regions based on a preset risk confidence calculation function, combined with the discharge risk distribution data and the image noise probability; and a defect labeling module for labeling candidate defect regions in the original image whose risk confidence scores are higher than a preset score threshold.

[0008] According to another aspect of this application, an electronic device is provided, including a memory and a processor, the memory being used to store computer-executable instructions, and the processor being used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method described above.

[0009] According to another aspect of this application, a computer storage medium is provided that stores computer-executable instructions thereon, which, when executed by a processor, implement the steps of the method described above.

[0010] Compared with the prior art, the deep learning-based image recognition system and method for insulation defects in power equipment according to the embodiments of this application can deeply integrate visual defect detection with electrical topology risk assessment. It can realize spatial and quantitative risk judgment of defects through discharge path risk heat map, and combine image noise analysis for credibility correction, thereby reducing false alarm and false alarm rates and realizing the differentiation and identification of high-risk defects. Attached Figure Description

[0011] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0012] Figure 1 This is a flowchart of the deep learning-based image recognition method for insulation defects in power equipment according to the present invention.

[0013] Figure 2 This is a block diagram of the deep learning-based image recognition system for insulation defects in power equipment according to the present invention.

[0014] Figure 3 This is a block diagram of an electronic device according to the present invention. Detailed Implementation

[0015] Hereinafter, exemplary embodiments according to this application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments of this application. It should be understood that this application is not limited to the exemplary embodiments described herein.

[0016] Application Overview

[0017] In existing technologies, the detection of insulation defects in power equipment mainly relies on image recognition technology, which uses deep learning models to analyze the surface texture features of the equipment to locate the defect area. However, this method only focuses on the identification of defect morphology and does not consider the impact of electrical topology on discharge risk, resulting in the detection results failing to accurately reflect the actual degree of danger of the defect.

[0018] To address the aforementioned issues, the inventors discovered that relying solely on image features is insufficient to meet risk assessment requirements. It is necessary to integrate electrical topology data with visual information. By analyzing the correlation between discharge phenomena and equipment structure, they proposed constructing a discharge path risk heat map by combining electrical parameters and spatial location. Furthermore, they introduced noise probability calculation to eliminate background interference, ultimately forming a defect assessment system based on multi-source data fusion.

[0019] Exemplary methods

[0020] Figure 1The illustration shows a deep learning-based image recognition method for insulation defects in power equipment according to an embodiment of this application, comprising: receiving an original image; dividing the original image into a foreground region, a midground region, and a background region, wherein the foreground region is the main body of the insulating equipment, the midground region is the electrical components connected to the main body, and the background region is the image portion unrelated to the main body; extracting electrical topology data corresponding to the main body, the electrical topology data including the electrical connection relationship and electrical parameters between the main body and the electrical components; constructing a discharge path risk heatmap corresponding to the space of the original image based on the electrical topology data and the midground region, the discharge path risk heatmap including the discharge probability distribution and spatial location between electrical nodes; performing target detection on the foreground region and the midground region using a deep learning-based target detection model to obtain candidate defect regions; extracting discharge risk distribution data for each candidate defect region based on the discharge path risk heatmap; extracting image feature data of the background region and calculating the image noise probability of the candidate defect regions based on the image feature data; calculating the risk confidence score of the candidate defect regions based on a preset risk confidence calculation function, combined with the discharge risk distribution data and the image noise probability; and marking candidate defect regions in the original image whose risk confidence scores are higher than a preset score threshold.

[0021] The foreground region refers to the image portion containing the main body of the insulating equipment. This region is divided by extracting the main body's contour using image segmentation algorithms, ensuring that subsequent analysis focuses on the core equipment. The midground region contains components electrically connected to the main body. Edge detection combined with morphological processing is used to identify the boundaries of these connected components, establishing the spatial relationships between electrical nodes. Electrical topology data includes the electrical connections and voltage parameters between devices, which can be obtained through device database matching or image feature analysis, providing a physical basis for calculating discharge probability.

[0022] The preset risk confidence calculation function is as follows: ,in, This represents the normalized discharge risk value of the discharge risk distribution data at the corresponding location in the discharge risk heatmap for the i-th candidate defect region. This represents the probability of image noise corresponding to the candidate defect region. This is a risk weighting coefficient used to adjust the relative contributions of discharge risk and noise correction to the final score. , , The range of values ​​is , Score the risk confidence level. A larger value indicates that the defective area is both in a high-risk zone and not easily affected by noise.

[0023] Through the above technical solution, this application can accurately identify insulation defects with high discharge risk, avoid over-maintenance of low-risk defects, and provide maintenance personnel with detection results with clear risk levels by quantitatively assessing the discharge probability and reliability of defect areas, thereby optimizing the allocation of maintenance resources. This method effectively solves the problem of insufficient risk assessment capability of traditional image recognition technology and enhances the practical application value of defect detection results.

[0024] This application further proposes that before marking candidate defect regions with risk confidence scores higher than a preset score threshold in the original image, the process includes: receiving infrared thermal imaging data corresponding to the original image; extracting thermal anomaly regions in the infrared thermal imaging data whose temperature gradient is greater than a preset temperature difference threshold, and generating a thermal anomaly distribution layer; superimposing the thermal anomaly distribution layer onto the discharge path risk heat map; and increasing the risk confidence score of candidate defect regions located within the intersection of the thermal anomaly distribution layer and the discharge risk heat map according to a preset first weighting coefficient.

[0025] Infrared thermal imaging data refers to image data reflecting the surface temperature distribution of equipment, collected by infrared sensors. Specifically, it can be acquired in real-time using a non-contact infrared thermal imager. This data is used to detect abnormal temperature rise areas caused by partial discharge or poor contact. The thermal anomaly distribution layer is a spatial mapping layer of abnormal temperature gradient regions filtered by setting a temperature difference threshold. It can be generated using image segmentation algorithms combined with temperature gradient calculations. This distribution layer can characterize potential thermal fault areas of the equipment. The preset first weighting coefficient is a weight adjustment parameter set based on the spatial correlation between thermal anomalies and discharge risk. It can be set through historical data statistics or expert experience to enhance the credibility of candidate regions that simultaneously exhibit thermal anomalies and high discharge risk.

[0026] Specifically, after obtaining the infrared thermal imaging data corresponding to the original image, the area with significant temperature changes on the device surface is determined by temperature gradient calculation. When the temperature gradient of a certain area exceeds a preset threshold, the area is marked as a thermal anomaly area and a corresponding spatial distribution layer is generated. After spatially superimposing this distribution layer with the previously constructed discharge path risk heat map, overlapping areas with both thermal anomalies and high discharge risk probabilities can be identified. For candidate defect areas located within this overlapping area, their risk confidence score is improved by a preset weighting coefficient, so that defect areas with both thermal anomaly characteristics and discharge risk characteristics receive higher priority.

[0027] Through the above technical solution, this application can effectively combine the thermal characteristics and electrical topology risk characteristics in the equipment operation state to improve the accuracy of identifying potential discharge defects. Through the dynamic adjustment mechanism of risk confidence score, the probability of misjudgment caused by environmental noise or image interference can be reduced, enabling maintenance personnel to prioritize the handling of high-risk defect areas with multiple risk characteristics.

[0028] This application further proposes that before marking candidate defect regions with risk confidence scores higher than a preset score threshold in the original image, the process includes: receiving partial discharge acoustic signal data corresponding to the original image; performing time-frequency analysis on the acoustic signal data to extract acoustic feature parameters; matching the acoustic feature parameters with the corresponding spatial positions of the candidate defect regions in the discharge path risk heat map according to a preset mapping rule to obtain acoustic signal feature values ​​corresponding to each candidate defect region; calculating a fusion score based on the acoustic signal feature values ​​and the discharge risk distribution data of the candidate defect regions according to a preset fusion function; and adjusting the risk confidence score of the candidate defect regions according to the comparison result between the fusion score and the preset threshold, specifically including: when the fusion score is higher than the threshold, increasing the risk confidence score of the candidate defect region according to a preset second weighting coefficient; otherwise, maintaining the risk confidence score.

[0029] Among them, partial discharge acoustic signal data refers to ultrasonic or audible sound wave signals generated by power equipment during operation, collected by acoustic sensors. Specifically, this can be achieved using piezoelectric sensor arrays to capture the physical phenomenon of partial discharge caused by insulation defects. Time-frequency analysis involves performing short-time Fourier transform or wavelet transform processing on the acoustic signal, specifically using Morlet wavelet basis functions to extract the frequency distribution characteristics of the signal within different time windows. Acoustic characteristic parameters include the dominant frequency band energy proportion, harmonic distortion rate, and pulse repetition frequency, which can be calculated through spectral energy integration to characterize the intensity and mode characteristics of the discharge signal. Preset mapping rules refer to establishing a geometric correspondence between the acoustic signal propagation path and the spatial location of the image, specifically implemented using sound source localization algorithms to spatially correlate acoustic features with the visual detection area.

[0030] Specifically, after acquiring candidate defect areas, acoustic signals from the device during operation are simultaneously collected. Time-frequency analysis is used to extract feature parameters including the proportion of energy in the main frequency band and harmonic distortion rate. A sound source localization algorithm is used to determine the spatial location of the acoustic signal in the image coordinate system, and the acoustic feature parameters are mapped to the corresponding candidate defect areas. For each candidate defect area, its discharge risk distribution data and matched acoustic feature values ​​are input into a fusion function. When the fusion score exceeds a preset threshold, it indicates that the area contains a high-risk signal with dual acoustic-electrical feature verification, and its risk confidence score is increased through a weighting coefficient. If the threshold is not reached, the original score is maintained to avoid misjudgments caused by single-modal data.

[0031] More specifically, the preset fusion function is: ,in, This represents the normalized discharge risk value of the discharge risk distribution data at the corresponding location in the discharge risk heatmap for the i-th candidate defect region. The normalized acoustic signal feature values ​​extracted from time-frequency acoustic wave analysis for this candidate defect region. To control the relative contribution of discharge risk and acoustic features to the fusion score, acoustic feature fusion weighting coefficients are used. , The range of values ​​is .

[0032] Through the above technical solution, this application can combine the acoustic characteristics of the equipment during operation with the visual inspection results to perform dual verification of the discharge risk of candidate defect areas. When the image detection results and the acoustic signal match in terms of spatial location and feature intensity, the risk confidence score is improved, thereby reducing the probability of misjudgment caused by image noise or single data deviation and improving the accuracy of high-risk defect identification.

[0033] This application further proposes the following steps for constructing a discharge path risk heatmap: extracting spatial relative position constraint information of each electrical component in the mid-field region; generating a set of discharge paths between several electrical node pairs based on the electrical connection relationships and spatial relative position constraint information in the electrical topology data; calculating the discharge risk probability of each discharge path in the discharge path set based on the electrical parameters in the electrical topology data using a pre-trained discharge risk calculation model; and mapping the discharge risk probability to the spatial coordinate region corresponding to the discharge path in the original image to generate a discharge path risk heatmap containing the spatial distribution of the discharge risk probability of each discharge path.

[0034] Among these, the spatial relative position constraint information refers to the geometric positional relationship of electrical components in the image. Specifically, this can be achieved by identifying the electrical component regions using image segmentation algorithms, obtaining the coordinates of the center points of each component through coordinate analysis, and then calculating the distance and azimuth between the components. The set of discharge paths between electrical node pairs refers to the potential discharge channels generated based on electrical connection relationships and spatial position constraints. Specifically, this can be achieved by using graph theory algorithms to traverse adjacent nodes in the topology and generating candidate paths by combining the shortest spatial distance between components. The pre-trained discharge risk calculation model refers to a machine learning model trained based on historical discharge data. Specifically, this can be achieved using a gradient boosting decision tree algorithm, taking electrical parameters as input and outputting a predicted path discharge probability value.

[0035] Specifically, when constructing a discharge path risk heatmap, the mid-field region is first segmented to identify the location information of electrical components such as conductor joints and insulator strings. Spatial topological relationships between components are established through structured modeling, such as calculating the Euclidean distance and azimuth between the center points of two adjacent insulators. Component pairs with direct electrical connections are selected based on their electrical connections, and possible discharge paths are generated by combining their spatial locations. Parameters such as the voltage level and creepage distance involved in the path are input into a pre-trained model to calculate the discharge probability of each path. Finally, the probability values ​​are mapped onto the image coordinate system to form a probability heatmap superimposed on the original image, where the color intensity represents the discharge risk level of different areas.

[0036] Through the above technical solution, this application can spatially correlate image detection results with electrical risk parameters, effectively distinguishing defect areas that are similar in appearance but have different risk levels.

[0037] This application further proposes to extract the spatial relative position constraint information of each electrical component in the mid-field region, including: performing image segmentation on the mid-field region to identify multiple electrical component regions; performing structured modeling on the electrical component regions to obtain the spatial position information of each electrical component in the original image; and calculating the relative distance and direction relationship between the center points of any two electrical components based on the spatial position information to generate spatial relative position constraint information.

[0038] Image segmentation refers to dividing a mid-range image into multiple independent electrical component regions. This can be achieved using deep learning-based semantic segmentation algorithms, such as the U-Net network structure, which identifies the contour boundaries of different electrical components through pixel-level classification. Structured modeling involves extracting geometric features from the segmented electrical component regions. This can be done by using edge detection combined with morphological operations to obtain the coordinates of the minimum bounding rectangle of each component, thereby establishing a structured data model containing positional and dimensional information. The relative distance and orientation relationship between center points is determined by calculating the Euclidean distance and the angle between the center points of two component regions. This can be done using coordinate difference calculations in a Cartesian coordinate system to generate vector parameters representing the spatial relationship between components.

[0039] Specifically, the process begins with semantic segmentation of the mid-field region of the input image. A pre-trained convolutional neural network model then identifies the precise contours of electrical components such as circuit breaker terminals and cable connectors. Next, geometric feature analysis is performed on each segmented region, extracting parameters such as the center point coordinates, length, and width of each component in the image coordinate system to form a structured location database. Further, for any two electrical component regions, the straight-line distance and azimuth angle between them are calculated using the center point coordinates. For example, the distance formula between two points and the arctangent function are used to obtain the distance and azimuth angle values, ultimately generating a set of constraint information containing the spatial relationships between all components.

[0040] Through the above technical solution, this application realizes the automated calibration of the spatial position of electrical components, and constructs a quantifiable spatial constraint system through vectorized distance and direction calculation. This provides accurate spatial position constraints for the subsequent construction of discharge path risk heat map, thereby improving the reliability of discharge risk probability calculation and the credibility of defect detection results.

[0041] This application further proposes to extract image feature data of background regions, including: dividing the background region into sub-regions to obtain several background sub-regions; calculating the brightness variance value and texture variability of each background sub-region; identifying background sub-regions with brightness variance values ​​greater than a preset brightness variance threshold and texture variability greater than a preset texture variability threshold as unstable background sub-regions; and extracting the brightness variance value and texture variability of the unstable background sub-regions as image feature data.

[0042] Sub-region segmentation refers to dividing the background region into multiple regular or irregular local regions. This can be achieved using grid partitioning or content-based adaptive segmentation algorithms, and is used for refined analysis of local features of the background region. Brightness variance refers to the dispersion of pixel brightness values ​​within a sub-region, specifically obtained by calculating the variance of pixel brightness values, and is used to reflect the brightness fluctuations of the background sub-region. Texture variability refers to the complexity of texture features within a sub-region, specifically achieved by calculating contrast or entropy using the gray-level co-occurrence matrix, and is used to characterize the dynamic changes in texture of the background sub-region.

[0043] Specifically, after the background region is divided into multiple sub-regions, the luminance variance and texture variability are calculated for each sub-region. Sub-regions with higher luminance variance values ​​indicate the presence of lighting changes or reflective interference, while sub-regions with higher texture variability may contain dynamic objects or complex backgrounds. By setting thresholds, unstable sub-regions that simultaneously meet the conditions of luminance variance and texture variability are selected, and their luminance variance and texture variability values ​​are extracted as image feature data. This data is subsequently used to evaluate whether candidate defect regions are affected by background noise. For example, when candidate defect regions and unstable background sub-regions have spatial overlap, their image noise probability will be significantly increased.

[0044] In the above scheme, for areas outside the unstable background sub-regions, since the brightness and texture changes are very small, they belong to the "normal and consistent" background part of the environment. The probability of image noise and artifact interference in these areas is low, and they have almost no negative impact on the identification of candidate defect areas. If all background sub-regions are considered, it will lead to a waste of computing resources. Therefore, this scheme only focuses on the unstable background sub-regions.

[0045] Through the above technical solution, this application can effectively distinguish between real defects and false detection areas caused by background noise, reduce the probability of misjudgment caused by changes in lighting, dynamic objects or complex textures, and improve the reliability of defect detection results.

[0046] This application further proposes to calculate the image noise probability of candidate defect regions based on image feature data, including: extracting the brightness variance and texture variation of unstable background sub-regions that overlap with the candidate defect regions in the original image; converting the brightness variance and texture variation into the image noise probability of the corresponding candidate defect regions based on a preset noise probability mapping model; and setting the image noise probability of candidate defect regions that do not overlap with the unstable background sub-regions to zero.

[0047] Among them, the noise probability mapping model refers to a mathematical model that converts brightness variance and texture variation into noise probability. Specifically, it can be implemented using a linear regression model or a probability lookup table based on historical data, and is used to establish the mapping relationship between background interference features and noise probability.

[0048] Specifically, after background region segmentation, unstable background sub-regions are selected based on brightness variance and texture variability. When a candidate defect region spatially overlaps with an unstable background sub-region, the brightness variance and texture variability of the corresponding sub-region are extracted and input into a noise probability mapping model to obtain the noise probability of that defect region. If a candidate defect region does not overlap with an unstable background sub-region, its noise probability is directly set to zero. This process quantifies the illumination fluctuations and texture complexity of the background region, correlates the noise interference probability of the candidate defect region with background stability, and thus effectively reduces the false detection probability caused by background interference in the risk confidence score.

[0049] Through the above technical solution, this application can effectively distinguish between false defect areas caused by background interference and real insulation defect areas, reduce the false detection rate caused by light fluctuations or complex background textures, and make the risk confidence score more accurately reflect the real discharge risk of the defect.

[0050] Exemplary System

[0051] Figure 2 The illustration shows the receipt of an original image according to an embodiment of this application; a region division module for dividing the original image into a foreground region, a midground region, and a background region, wherein the foreground region is the main body of the insulating equipment, the midground region is the electrical components connected to the main body, and the background region is the image portion unrelated to the main body; an electrical topology extraction module for extracting electrical topology data corresponding to the main body, the electrical topology data including electrical connection information between the main body and electrical components; a risk heatmap construction module for constructing a discharge path risk heatmap corresponding to the space of the original image based on the electrical topology data and the midground region, the discharge path risk heatmap including the discharge probability distribution and spatial location between electrical nodes; and a target detection module. The system comprises the following modules: a target detection module for detecting targets in the foreground and midground regions using a deep learning-based target detection model to obtain candidate defect regions; a risk mapping module for extracting discharge risk distribution data for each candidate defect region based on the discharge path risk heatmap; an image feature extraction module for extracting image feature information from the background region and calculating the image noise probability of the candidate defect regions based on the image feature data; a risk confidence calculation module for calculating the risk confidence score of the candidate defect regions based on a preset risk confidence calculation function, combined with discharge risk distribution data and image noise probability; and a defect annotation module for annotating candidate defect regions in the original image whose risk confidence scores are higher than a preset score threshold.

[0052] In one example, the risk heatmap construction module constructs a discharge path risk heatmap by: extracting the spatial relative position constraint information of each electrical component in the mid-field region; generating a set of discharge paths between several electrical node pairs based on the electrical connection relationships and spatial relative position constraint information in the electrical topology data; calculating the discharge risk probability of each discharge path in the discharge path set based on the electrical parameters in the electrical topology data using a pre-trained discharge risk calculation model; and mapping the discharge risk probability to the spatial coordinate region corresponding to the discharge path in the original image to generate a discharge path risk heatmap containing the spatial distribution of the discharge risk probability of each discharge path.

[0053] In one example, the risk heatmap construction module extracts the spatial relative position constraint information of each electrical component in the mid-field area by: performing image segmentation on the mid-field area to identify multiple electrical component regions; performing structured modeling on the electrical component regions to obtain the spatial position information of each electrical component in the original image; and calculating the relative distance and direction relationship between the center points of any two electrical components based on the spatial position information to generate spatial relative position constraint information.

[0054] In one example, the image feature extraction module extracts image feature data of the background region by: dividing the background region into sub-regions to obtain several background sub-regions; calculating the brightness variance value and texture variability of each background sub-region; identifying background sub-regions with brightness variance values ​​greater than a preset brightness variance threshold and texture variability values ​​greater than a preset texture variability threshold as unstable background sub-regions; and extracting the brightness variance value and texture variability of the unstable background sub-regions as image feature data.

[0055] In one example, the image feature extraction module calculates the image noise probability of the candidate defect region based on the image feature data by: extracting the brightness variance and texture variability of the unstable background sub-region that overlaps with the candidate defect region in the original image; converting the brightness variance and texture variability into the image noise probability of the corresponding candidate defect region based on a preset noise probability mapping model; and setting the image noise probability of the candidate defect region that does not overlap with the unstable background sub-region to zero.

[0056] Exemplary electronic devices

[0057] Figure 3 An electronic device according to an embodiment of this application is illustrated. The electronic device may be the mobile device itself, or a standalone device independent of it, which may communicate with the mobile device to receive collected input signals from it and send selected target driving behaviors to it.

[0058] Figure 3 A block diagram of an electronic device according to an embodiment of this application is illustrated.

[0059] like Figure 3 As shown, the electronic device includes one or more processors and memory.

[0060] A processor can be a central processing unit (CPU) or other form of processing unit with data processing and / or instruction execution capabilities, and can control other components in an electronic device to perform desired functions.

[0061] The memory may include one or more computer program products, which may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and a processor may execute the program instructions to implement the driving behavior decision-making methods of the various embodiments of this application described above, and / or other desired functions.

[0062] In one example, the electronic device may also include input devices and output devices, which are interconnected via a bus system and / or other forms of connection mechanism (not shown).

[0063] Of course, for the sake of simplicity, Figure 3 Only some of the components of the electronic device relevant to this application are shown, omitting components such as buses, input / output interfaces, etc. In addition, the electronic device may include any other suitable components depending on the specific application.

[0064] Exemplary computer-readable media

[0065] Embodiments of this application may also be computer-readable storage media storing computer program instructions thereon, which, when executed by a processor, cause the processor to perform the steps in the driving behavior decision-making methods according to various embodiments of this application described in the "Exemplary Methods" section above.

[0066] Computer-readable storage media may take the form of any combination of one or more readable media. A readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may, for example, include, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: electrical connections having one or more wires, portable 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 thereof.

[0067] The basic principles of this application have been described above with reference to specific embodiments. However, it should be noted that the advantages, benefits, and effects mentioned in this application are merely examples and not limitations, and should not be considered as essential features of each embodiment of this application. Furthermore, the specific details disclosed above are for illustrative and facilitative purposes only, and are not limitations. These details do not limit the application to the necessity of employing the aforementioned specific details for implementation.

[0068] The block diagrams of devices, apparatuses, devices, and systems involved in this application are merely illustrative examples and are not intended to require or imply that they must be connected, arranged, or configured in the manner shown in the block diagrams. As those skilled in the art will recognize, these devices, apparatuses, devices, and systems can be connected, arranged, and configured in any manner. Words such as “comprising,” “including,” “having,” etc., are open-ended terms meaning “including but not limited to,” and are used interchangeably with them. The terms “or” and “and” as used herein refer to the terms “and / or,” and are used interchangeably with them unless the context clearly indicates otherwise. The term “such as” as used herein refers to the phrase “such as but not limited to,” and is used interchangeably with it.

[0069] It should also be noted that in the apparatus, equipment, and methods of this application, the components or steps can be disassembled and / or recombined. These disassemblies and / or recombinations should be considered as equivalent solutions of this application.

[0070] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use this application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein can be applied to other aspects without departing from the scope of this application. Therefore, this application is not intended to be limited to the aspects shown herein, but rather to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0071] The above description has been given for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of this application to the forms disclosed herein. Although numerous exemplary aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, alterations, additions, and sub-combinations thereof.

Claims

1. A deep learning-based image recognition method for insulation defects in power equipment, characterized in that, include: Receive the raw image; The original image is divided into a foreground region, a midground region, and a background region. The foreground region is the main body of the insulating equipment, the midground region is the electrical components connected to the main body, and the background region is the image portion unrelated to the main body. Extract electrical topology data corresponding to the main body, wherein the electrical topology data includes the electrical connection relationship and electrical parameters between the main body and the electrical components; Based on the electrical topology data and the mid-field region, a discharge path risk heatmap corresponding to the original image space is constructed. The discharge path risk heatmap includes the discharge probability distribution and spatial location between electrical nodes. The construction of the discharge path risk heatmap includes: extracting the spatial relative position constraint information of each electrical component in the mid-field region; generating a set of discharge paths between several pairs of electrical nodes based on the electrical connection relationships in the electrical topology data and the spatial relative position constraint information; calculating the discharge risk probability of each discharge path in the discharge path set based on the electrical parameters in the electrical topology data using a pre-trained discharge risk calculation model; and mapping the discharge risk probability to the spatial coordinate region corresponding to the discharge path in the original image to generate a discharge path risk heatmap containing the spatial distribution of the discharge risk probability of each discharge path. Candidate defect regions are obtained by performing target detection on the foreground region and the midground region using a deep learning-based target detection model. Based on the discharge path risk heatmap, the discharge risk distribution data of each candidate defect region is extracted; Extract image feature data of the background region, and calculate the image noise probability of the candidate defect region based on the image feature data; Based on a preset risk confidence calculation function, and combining the discharge risk distribution data with the image noise probability, the risk confidence score of the candidate defect region is calculated. Candidate defect areas with risk confidence scores higher than a preset score threshold are marked in the original image.

2. The deep learning-based image recognition method for insulation defects in power equipment according to claim 1, characterized in that, Before marking candidate defect regions in the original image whose risk confidence scores are higher than a preset scoring threshold, the method further includes: Receive infrared thermal imaging data corresponding to the original image; Extract thermal anomaly regions from the infrared thermal imaging data whose temperature gradient is greater than a preset temperature difference threshold, and generate a thermal anomaly distribution layer. The thermal anomaly distribution layer is superimposed onto the discharge path risk heat map; The risk confidence score of candidate defect regions located within the intersection of the thermal anomaly distribution layer and the discharge risk heatmap is increased based on a preset first weighting coefficient.

3. The deep learning-based image recognition method for insulation defects in power equipment according to claim 1 or 2, characterized in that, Before marking candidate defect regions in the original image whose risk confidence scores are higher than a preset scoring threshold, the method further includes: Receive partial discharge acoustic signal data corresponding to the original image; Time-frequency analysis is performed on the acoustic signal data to extract acoustic characteristic parameters; The acoustic wave feature parameters are matched with the corresponding spatial positions of the candidate defect regions in the discharge path risk heat map according to a preset mapping rule to obtain the acoustic wave signal feature values ​​corresponding to each candidate defect region. Based on the acoustic signal feature values ​​and the discharge risk distribution data of the candidate defect region, a fusion score is calculated according to a preset fusion function. Based on the comparison between the fusion score and the preset threshold, the risk confidence score of the candidate defect region is adjusted. Specifically, when the fusion score is higher than the preset threshold, the risk confidence score of the candidate defect region is increased according to a preset second weighting coefficient; otherwise, the risk confidence score is maintained.

4. The deep learning-based image recognition method for insulation defects in power equipment according to claim 1, characterized in that, The extraction of spatial relative position constraint information for each electrical component in the mid-field region includes: Image segmentation was performed on the mid-field region to identify multiple electrical component areas; The electrical component region is structured and modeled to obtain the spatial location information of each electrical component in the original image; The spatial relative position constraint information is generated by calculating the relative distance and direction relationship between the center points of any two electrical components based on the spatial position information.

5. The deep learning-based image recognition method for insulation defects in power equipment according to claim 1, characterized in that, The extraction of image feature data of the background region includes: The background region is divided into sub-regions to obtain several background sub-regions; Calculate the luminance variance and texture variation for each of the background sub-regions; Background sub-regions whose brightness variance value is greater than a preset brightness variance threshold and whose texture variability is greater than a preset texture variability threshold are defined as unstable background sub-regions. The brightness variance and texture variation values ​​in the unstable background sub-region are extracted as the image feature data.

6. The deep learning-based image recognition method for insulation defects in power equipment according to claim 5, characterized in that, The step of calculating the image noise probability of the candidate defect region based on the image feature data includes: Extract the brightness variance and texture variation of unstable background sub-regions that spatially overlap with the candidate defect regions in the original image; Based on a preset noise probability mapping model, the brightness variance value and texture variation degree are converted into the image noise probability of the corresponding candidate defect region; Set the image noise probability of candidate defect regions that do not spatially overlap with the unstable background sub-region to zero.

7. A deep learning-based image recognition system for insulation defects in power equipment, characterized in that, include: The image receiving module is used to receive raw images; The region segmentation module is used to divide the original image into a foreground region, a midground region, and a background region. The foreground region is the main body of the insulating equipment, the midground region is the electrical components connected to the main body, and the background region is the image portion unrelated to the main body. An electrical topology extraction module is used to extract electrical topology data corresponding to the main body. The electrical topology data includes electrical connection information and electrical parameters between the main body and the electrical components. A risk heatmap construction module is used to construct a discharge path risk heatmap corresponding to the space of the original image based on the electrical topology data and the mid-field region. The discharge path risk heatmap includes the discharge probability distribution and spatial location between electrical nodes. The construction of the discharge path risk heatmap includes: extracting the spatial relative position constraint information of each electrical component in the mid-field region; generating a set of discharge paths between several pairs of electrical nodes based on the electrical connection relationships in the electrical topology data and the spatial relative position constraint information; calculating the discharge risk probability of each discharge path in the discharge path set based on the electrical parameters in the electrical topology data using a pre-trained discharge risk calculation model; and mapping the discharge risk probability to the spatial coordinate region corresponding to the discharge path in the original image to generate a discharge path risk heatmap containing the spatial distribution of the discharge risk probability of each discharge path. The target detection module is used to perform target detection on the foreground region and the midground region using a deep learning-based target detection model to obtain candidate defect regions; The risk mapping module is used to extract the discharge risk distribution data of each of the candidate defect regions based on the discharge path risk heat map; The image feature extraction module is used to extract image feature data of the background region and calculate the image noise probability of the candidate defect region based on the image feature data. The risk confidence calculation module is used to calculate the risk confidence score of the candidate defect region based on a preset risk confidence calculation function, combined with the discharge risk distribution data and the image noise probability; The defect annotation module is used to annotate candidate defect regions in the original image where the risk confidence score is higher than a preset score threshold.

8. An electronic device comprising a memory and a processor, characterized in that: The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the method as described in any one of claims 1 to 6.

9. A computer storage medium storing computer-executable instructions thereon, characterized in that: When the computer-executable instructions are executed by a processor, they implement the steps of the method as described in any one of claims 1 to 6.

Images