Composite insulator large cluster parachute heating defect judgment interference suppression method

Abstract

The invention discloses a composite insulator large cluster parachute heating defect judgment interference suppression method. The system comprises an infrared image acquisition and preprocessing module, a composite insulator target detection and positioning module, a composite insulator temperature field analysis module, a large umbrella group reflection area identification module, a composite insulator central axis reconstruction module and a heating defect intelligent diagnosis and interference suppression module. Complex infrared image analysis is converted into refined analysis of a temperature field and geometric features, and meanwhile, a strategy of combining deep learning and morphology is introduced to improve the precision of target detection and segmentation and ensure the robustness of central axis reconstruction and interference suppression. The method has the beneficial effects that the false alarm rate can be obviously reduced, the detection accuracy of the real heating defect is improved, the result is highly consistent with the interpretation result of an artificial expert, the processing efficiency is high, and the method is suitable for a large-scale unmanned aerial vehicle inspection scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent defect detection of power grid power transmission equipment, in particular to a composite insulator large umbrella group heat defect judgment interference suppression method. BACKGROUND

[0002] In recent years, with the continuous expansion of the scale of China's power grid and the continuous improvement of voltage level, the safe and stable operation of the power transmission line is increasingly important to the national economy. As a key insulating device of modern power transmission lines, composite insulators are widely used in various overhead lines due to their light weight, high mechanical strength, good anti-pollution flashover performance and other advantages. However, in the long-term operation process, composite insulators will abnormally heat due to internal defects (such as core rod brittle fracture, breakdown) and external factors (surface contamination, electric arc), and this heat is an important early sign of insulator performance degradation. If not discovered and handled in time, minor defects may develop into serious faults such as breakdown and fracture, ultimately causing line tripping and even power grid splitting and other major accidents, threatening the reliability of power supply.

[0003] Under this background, infrared thermal imaging detection technology has become the mainstream means of composite insulator state detection and fault diagnosis due to its non-contact, intuitive and efficient advantages. In particular, with the maturity of unmanned aerial vehicle technology, unmanned aerial vehicles equipped with infrared cameras for inspection have realized fast and flexible temperature survey of high-altitude line equipment, significantly improving the operation and maintenance efficiency. At present, the automatic identification of composite insulator heat defects based on unmanned aerial vehicle infrared images usually relies on temperature analysis of the central axis region of the insulator. The basic principle of this method is that the temperature distribution of a normal insulator should show uniform or smooth transition characteristics along the central axis; if there are abnormal high temperature points or significant temperature distribution gradient changes in the local area, it indicates that there may be defects at that location. This judgment logic based on the temperature characteristics of the center line has been proven to have good indication effect on most heat defects in practice.

[0004] However, the existing center line automatic identification method has obvious limitations when dealing with composite insulators with special umbrella skirt structures, especially for composite insulator models equipped with large umbrella skirts. The design intention of the large umbrella group is to provide better rain flash and anti-pollution flash performance in bad weather, but the material surface characteristics and special curved geometric design make it extremely easy to produce strong mirror reflection effect under sunlight. This reflection is not the heat of the insulator itself, but it will form a high-light area similar to the real heat in the infrared image, and its temperature characteristics are easily confused with the real overheating caused by internal defects or local electric arcs of the core rod.

[0005] This "pseudo-heat" interference caused by reflection brings severe challenges to the automatic diagnosis based on infrared images. On the one hand, it leads to a significant increase in the false positive rate of the defect identification system, and the operation and maintenance personnel need to invest a lot of effort in manually reviewing the alarms in the massive inspection pictures, which greatly offsets the efficiency advantage of unmanned aerial vehicle inspection and increases unnecessary operation and maintenance burden. On the other hand, simply increasing the alarm temperature or threshold to reduce the false positive rate may lead to missed reports of real, especially early and weak heat defects, which poses a safety hazard. Therefore, how to effectively distinguish the temperature interference caused by the reflection of large umbrella group from the temperature characteristics of real heat defects, that is, to realize reliable interference suppression, has become a key technical bottleneck for improving the accuracy and practicality of infrared detection of composite insulators, and is also a practical problem that needs to be solved in the field of intelligent inspection of power equipment. SUMMARY

[0006] The present application is to overcome the above-mentioned deficiencies in the prior art, to better realize the interference suppression of the heat of the large umbrella group of the composite insulator, and provides a composite insulator large umbrella group heat defect judgment interference suppression method for realizing correct and rapid detection of heat defects of the composite insulator of the power transmission line.

[0007] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions: A composite insulator large umbrella group heat defect judgment interference suppression method, comprising an infrared image acquisition and preprocessing module, a composite insulator target detection and positioning module, a composite insulator temperature field analysis module, a large umbrella group reflection area identification module, a composite insulator center axis reconstruction module, and a heat defect intelligent diagnosis and interference suppression module, and the specific processing flow is as follows: (1) The infrared image acquisition and preprocessing module realizes the acquisition of infrared images through the preset OBS communication interface and the data interaction with the unmanned aerial vehicle inspection platform, and at the same time, the preprocessing process includes strict access verification of the input image, screening based on pixel resolution, and eliminating image data that does not meet the format requirements; by analyzing the image metadata, verifying its source, and ensuring the standardization of the input data; (2) The composite insulator target detection and positioning module uses a deep learning-based instance segmentation algorithm to analyze the input image through image recognition algorithm, and outputs the results from the detection head and the segmentation head, to generate the minimum bounding box of the composite insulator; (3) The composite insulator temperature field analysis module uses the pixel coordinates of the insulator segmentation area determined by the composite insulator target detection and positioning module as an index to extract the temperature at the corresponding position from the global radiation temperature matrix returned by the SDK, and form a corresponding temperature matrix; (4) The "big umbrella group" reflection area recognition module identifies the "big umbrella group" structure with a size significantly larger than the standard umbrella skirt through morphological analysis of the insulator segmentation area, and then analyzes the temperature distribution characteristics of these areas: focus on identifying areas that present uniform high brightness, with a gentle temperature gradient change, and a sudden change in temperature from the surrounding umbrella skirt. Finally, each identified "big umbrella group" area is positioned as a compact rectangular area, and its coordinates are recorded to provide spatial positioning for subsequent interference elimination. (5) The composite insulator center axis reconstruction module removes all the "big umbrella group" rectangular areas identified by the previous module from the original insulator segmentation mask to form an intermediate mask containing "holes". Then, using an image inpainting algorithm, the missing areas are smoothly interpolated to reconstruct a complete insulator contour with no "big umbrella group". On this basis, a skeletonization algorithm is used to extract the topological skeleton of the repaired contour as the center axis of the insulator, and the corresponding temperature sequence is extracted from the temperature field matrix based on the axis coordinates. (6) The heat defect intelligent diagnosis and interference suppression module uses a well-trained one-dimensional convolutional neural network to judge the heat of the center axis temperature sequence.

[0008] The present application solves the problem of "big umbrella group" reflection interference by using multiple modules for collaborative processing, converting complex infrared image analysis into fine analysis of temperature field and geometric features. By introducing a combination of deep learning and morphological analysis, the accuracy of target detection and segmentation is improved, and the robustness of center axis reconstruction and interference suppression is ensured. This method can significantly reduce the false positive rate and improve the accuracy of detecting real heat defects. The results are highly consistent with those of artificial expert interpretation, and the processing efficiency is high, making it suitable for large-scale unmanned aerial vehicle inspection scenarios. The present application provides reliable technical support for intelligent operation and maintenance of composite insulators for power transmission lines, and significantly improves the automation level of safe operation of power systems.

[0009] As a preferred embodiment, in step (1), the infrared image acquisition and preprocessing module is used to obtain raw data from distributed data sources through a standardized pipeline. The module interacts with the unmanned aerial vehicle inspection platform through a pre-set OBS communication interface to achieve batch acquisition of infrared images. The module executes a standardized preprocessing and quality evaluation pipeline, which first performs strict access verification on the input image: based on fixed pixel resolution, it automatically removes image data that does not meet the format requirements; then, by analyzing the image file metadata, it verifies the source of the acquisition equipment, thereby ensuring the standardization of the input data and the reliability of the subsequent temperature analysis process.

[0010] As preferred, in step (2), the composite insulator target detection and positioning module utilizes a fully trained deep neural network to perform multi-scale feature extraction and fusion based on a backbone network combined with a feature pyramid network; in the detection process, candidate target regions are first generated through a region extraction protocol and its anchor box mechanism, then the classification network is used to determine whether the region is a composite insulator, and the accurate bounding box coordinates are regressed, and subsequently, the pixel-level classification is performed on each confirmed positive candidate box by a further segmentation head to generate a high-precision binary segmentation mask, wherein the segmentation mask is represented by a matrix M, M(x, y) = 1 indicates that the pixel point (x, y) belongs to the composite insulator target, and M(x, y) = 0 indicates that it belongs to the background; by applying the non-maximum suppression algorithm, the redundant and overlapping detection results are removed according to the bounding box confidence and segmentation overlap, and finally the detected composite insulator in each image is output, which can tightly wrap the minimum rotating bounding rectangle of the composite insulator in any direction, and is accurately described by a five-tuple: wherein, represents the coordinates of the center point of the minimum rotating bounding rectangle in the image pixel coordinate system, w represents the length of the composite insulator rectangular frame in its own coordinate system, and h represents the width of the composite insulator rectangular frame in its own coordinate system. represents the rotation angle of the rectangular frame, i.e., the inclination angle of the composite insulator.

[0011] As preferred, in step (3), the composite insulator temperature field analysis module extracts the absolute temperature values corresponding to the composite insulator pixel region from the original infrared data based on the segmentation mask output by the composite insulator target detection and positioning module, and constructs a local temperature matrix dedicated to the composite insulator, and through the deep integration and calling of the SDK, the global radiation temperature matrix covering the entire infrared image field of view is obtained through the temperature interface , wherein R represents the infrared image pixel matrix, H and W represent the image height and width respectively, and each element (i, j) of the global temperature matrix stores the unprocessed original temperature value corresponding to the pixel coordinate (i, j) in the image.

[0012] As preferred, in step (3), specifically, the composite insulator temperature field analysis module takes the binary segmentation mask matrix generated by the composite insulator target detection and positioning module as a spatial index, in which the pixel points with a value of 1 constitute the accurate contour of the composite insulator, and through pixel-by-pixel mapping, the temperature values belonging only to the minimum rotating bounding matrix are extracted from the global temperature matrix , and these temperature values construct a local temperature matrix , .

[0013] As preferred, in step (4), the large umbrella group reflection region identification module realizes accurate identification based on the physical characteristics of the reflection region having set symmetry and uniformity of temperature distribution, and the specific process is as follows: (41) Based on the composite insulator rotating minimum circumscribed rectangle OBB parameters obtained by the composite insulator temperature field analysis module, the geometric center line parallel to the long is extracted; let the center point of the OBB be , the rotation angle be , the length be L, and the width be W, then the center line can be expressed as the equation: Sampling along the center line at a pixel interval, an ordered point set is obtained: Wherein, ; t represents a pixel distance of moving from the center point (x c , y c ) along the center line direction (cosθ, sinθ); i represents the pixel point on the center line; (42) For each center point , the following operations are performed: first, a straight line perpendicular to the center line is passed through the point, and the direction vector of the straight line is ; second, the intersection of the perpendicular line and the OBB boundary is determined to determine the sampling line segment Li; then the temperature values of all pixel points on the line segment Li are extracted to form the temperature profile Ti(s); the temperature similarity index of the temperature profile is calculated, and the center point temperature is set as Tc=Ti(0); if the temperature value of the temperature profile is , it is considered that the point is close to the center temperature, and the number of similar pixels on the temperature profile N same is counted; the similar points on the temperature profile are compared with all the points, and the similarity is ; if the number of similar points exceeds two-thirds, i.e. , it is considered that the perpendicular line is a candidate reflection line; wherein N total represents the number of all pixels on the temperature profile; (43) After scanning all the center lines, continuity analysis is performed; let the continuous candidate reflection line index set be {i, i+1,..., i+k}, if the continuous length k+1, it is determined that the region is a candidate reflection region; (44) In order to accurately locate the reflection region, the maximum value of the similarity in the continuous candidate line is found: A corresponding vertical line Lmax is determined as the symmetry axis of the large umbrella group; (45) Taking the vertical line Lmax as the symmetry axis of the large umbrella group, each L / 2 distance is verified along the center line direction, and the final large umbrella group region is defined as , and the module outputs all identified large umbrella group regions on the composite insulator .

[0014] As preferred, in step (5), the composite insulator center axis reconstruction module reconstructs the topological center axis reflecting the geometric characteristics of the insulator body structure after excluding the large umbrella group reflection interference region, and the process is as follows: (51) Perform large umbrella group interference region elimination, and delete all regions output by the large umbrella group reflection region identification module from the original composite insulator binary segmentation mask matrix M; through this step, the pixel value of the corresponding large umbrella group region in the mask is changed from 1 to 0, thereby forming a segmentation mask matrix M holey without the large umbrella group region; (52) Perform temperature constraint-based morphological repair to obtain a segmentation mask matrix M holey without the large umbrella group, use the average temperature of the pixels on the center line of the whole composite insulator as the filling value, and finally obtain M repaired after repair and elimination of the large umbrella group; (53) After obtaining the complete contour M repaired after elimination of the large umbrella group and repair, the previously obtained geometric center line pixel coordinates of the minimum circumscribed rectangle of the composite insulator are used again to extract the corresponding temperature sequence from the temperature matrix after elimination of the large umbrella group and repair filling in step (52), and the one-dimensional convolutional neural network is used to judge whether the composite insulator corresponding to the temperature sequence is heated.

[0015] As preferred, in step (6), the one-dimensional convolutional neural network is trained in advance using labeled heated and normal samples to learn the weak abnormal pattern in the temperature sequence; after inputting the temperature sequence, the network extracts local features through the convolutional layer, reduces the dimension through the pooling layer, and outputs the heating probability through the fully connected layer; combined with the threshold value, the final heating defect diagnosis result is output, and false positives caused by large umbrella group reflection are automatically suppressed, realizing interference suppression.

[0016] The beneficial effects of the present application are: it can significantly reduce the false positive rate, improve the detection accuracy of real heating defects, and is highly consistent with the results of artificial expert interpretation, has high processing efficiency, and is suitable for large-scale unmanned aerial vehicle inspection scenes; provides reliable technical support for intelligent operation and maintenance of composite insulators of power transmission lines, and effectively improves the automation level of safe operation of the power system. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 is a method flowchart of the present application.

[0018] Figure 2 is a whole architecture diagram of the present application. DETAILED DESCRIPTION

[0019] The present application is further described below in conjunction with the accompanying drawings and specific embodiments.

[0020] As Figure 1 , Figure 2 In the embodiment described in the foregoing, a composite insulator large umbrella group heat defect judgment interference suppression method comprises an infrared image acquisition and preprocessing module, a composite insulator target detection and positioning module, a composite insulator temperature field analysis module, a large umbrella group reflection area identification module, a composite insulator center axis reconstruction module, and a heat defect intelligent diagnosis and interference suppression module. The specific processing procedure is as follows: (1) The infrared image acquisition and preprocessing module realizes the acquisition of infrared images through a preset OBS communication interface and data interaction with a UAV inspection platform, and at the same time, the preprocessing procedure includes strict access verification on the input images, screening based on the pixel resolution (fixed as 640*512), and eliminating image data that does not meet the format requirements; by analyzing the image metadata, it is verified whether the source is a compatible DJI infrared camera platform, to ensure the standardization of the input data.

[0021] The infrared image acquisition and preprocessing module constitutes the data input and quality control of the whole system. A standardized pipeline is used to obtain the original from the distributed data source, and the module realizes the batch acquisition of transmission line infrared images through data interaction with the UAV inspection platform through a preset OBS communication interface. The module executes a standardized preprocessing and quality evaluation pipeline, first performs strict access verification on the input images: based on the fixed pixel resolution (640*512), performs primary automatic screening, and automatically eliminates image data that does not meet the format requirements (for example: resolution inconsistency, format error or damaged images); further, by analyzing the image file metadata, it is verified whether the source of the acquisition device is a compatible DJI infrared camera platform, so as to ensure the standardization of the input data and the reliability of the subsequent temperature analysis process. For images that meet the requirements, the module performs standardized preprocessing, including image denoising and contrast enhancement, to improve the accuracy of subsequent analysis.

[0022] (2) The composite insulator target detection and positioning module uses a deep learning-based instance segmentation algorithm (such as Mask R-CNN or YOLOv8-Seg) to analyze the preprocessed input image through image recognition algorithm, and outputs the results by the detection head and segmentation head, generating the minimum bounding box of the composite insulator.

[0023] The composite insulator target detection and positioning module accurately and robustly identifies the composite insulator from the complex UAV infrared background. The module uses a well-trained deep neural network (Mask R-CNN or YOLOv8-seg) to perform multi-scale feature extraction and fusion based on the backbone network combined with the feature pyramid network (FPN), which can effectively deal with the problem of inconsistent composite insulator scales caused by changes in shooting distance and angle in infrared images. During the detection process, the module first generates a large number of candidate target regions through the region proposal network (RPN) and its anchor box mechanism, then judges whether the region is a composite insulator through the classification network, and regresses the accurate bounding box coordinates. Subsequently, the segmentation head (Segmentation Head) further classifies each confirmed positive candidate box at the pixel level, generating a high-precision binary segmentation mask (Binary Mask). The segmentation mask is represented by a matrix M, where M(x,y)=1 indicates that the pixel point (x,y) belongs to the composite insulator target, and M(x,y)=0 indicates that it belongs to the background. The module applies the non-maximum suppression (NMS) algorithm to remove redundant and overlapping detection results according to the bounding box confidence and segmentation overlap, and finally outputs the detected composite insulator in each image. This composite insulator can tightly wrap the minimum rotating bounding box (OBB) of the inclined or arbitrary direction composite insulator, which is accurately described by a five-tuple: where, represents the coordinates of the center point of the minimum rotating bounding box in the image pixel coordinate system, w represents the length of the composite insulator rectangular frame in its own coordinate system (i.e. the length of the composite insulator), h represents the width of the composite insulator rectangular frame in its own coordinate system (i.e. the width of the composite insulator), represents the rotation angle of the rectangular frame, i.e. the inclination angle of the composite insulator, which provides a precise positioning basis for the subsequent modules.

[0024] (3) The composite insulator temperature field analysis module extracts the temperature at the corresponding position from the global radiation temperature matrix returned by the SDK by using the pixel coordinates of the insulator segmentation region determined by the composite insulator target detection and positioning module as the index, and forms the corresponding temperature matrix.

[0025] The composite insulator temperature field analysis module extracts the absolute temperature values ​​corresponding to the pixel regions of the composite insulator from the raw infrared data based on the precise segmentation mask M output by the composite insulator target detection and localization module, and constructs its dedicated local temperature matrix. Through deep integration and invocation of the DJI Infrared Development Kit (SDK), the module obtains a global radiation temperature matrix covering the entire infrared image field of view via a temperature interface. Where R represents the infrared image pixel matrix, and H and W represent the image height and width, respectively. Global temperature matrix. Each element (i,j) stores the unprocessed raw temperature value corresponding to pixel coordinates (i,j) in the image.

[0026] The composite insulator temperature field analysis module uses the binary segmentation mask matrix generated by the composite insulator target detection and localization module. As a spatial index, pixels with a value of 1 in this matrix constitute the precise outline of the composite insulator. The module obtains this information from the global temperature matrix through pixel-by-pixel mapping. Extract the temperature values ​​that belong only to the minimum rotated circumscribed matrix. These temperature values ​​construct a local temperature matrix. ,

[0027] This matrix contains only temperature data for the composite insulator region, ensuring the focus of subsequent analysis.

[0028] (4) The large umbrella cluster reflection area identification module identifies "large umbrella cluster" structures with significantly larger dimensions than standard umbrella skirts by performing morphological analysis on the segmented regions of the insulator. Then, it analyzes the temperature distribution characteristics of these regions: focusing on identifying areas that exhibit uniform high brightness, have a gentle temperature gradient, and show abrupt temperature changes compared to the surrounding umbrella skirts. Finally, each identified large umbrella cluster region is located as a compact rectangular region of interest (ROI), and its coordinates are recorded to provide spatial location for subsequent interference elimination.

[0029] The large umbrella cluster reflection area identification module accurately identifies pseudo-heating areas caused by solar radiation reflected from the surface of the composite insulator's large umbrella cluster. This module achieves accurate identification based on the physical characteristics of the reflection area's aggregate symmetry and uniform temperature distribution. The specific process is as follows: (41) Based on the parameters of the minimum circumscribed rectangle (OBB) of the composite insulator obtained from the composite insulator temperature field analysis module, extract its geometric center line parallel to the length. Let the center point of the OBB be... The rotation angle is If the length is L and the width is W, then the centerline can be represented by the equation: Sampling along the centerline with pixel interval, we get ordered point set: where, ; t represents a pixel distance from the center point (x c , y c ) along the centerline direction (cosθ, sinθ); i represents the pixel point on the centerline.

[0030] (42) For each center point , the relevant operations are as follows: first, pass through the straight line perpendicular to the centerline at the point, whose direction vector is ; second, determine the sampling line segment Li by the intersection of the perpendicular line and the OBB boundary; then extract the temperature values of all pixel points on the line segment Li to form the temperature profile Ti(s).

[0031] Calculate the temperature similarity index of this temperature profile. Let the center point temperature be Tc=Ti(0). If the temperature value of the temperature profile is , it is considered that the point is close to the center temperature, and the number of similar pixels on this temperature profile N same is counted. Compare the similar points on this temperature profile with all the points, and the similarity is . If the number of similar points exceeds two-thirds, i.e. , it is considered that this perpendicular line is a candidate reflection line; where N total represents the number of all pixels on the temperature profile.

[0032] (43) After scanning the entire centerline, perform continuity analysis. Let the continuous candidate reflection line index set be {i, i+1,..., i+k}. If the continuous length k+1 (indicates that the continuous candidate reflection line exceeds 6), it is determined that this region is a candidate reflection region.

[0033] (44) In order to accurately locate the reflection region, find the maximum similarity value in the continuous candidate line: Cluster the continuous candidate reflection lines, and take the perpendicular line corresponding to the maximum similarity value as the symmetry axis of the large umbrella group, i.e. the corresponding perpendicular line Lmax is determined as the symmetry axis of the large umbrella group.

[0034] (45) Take the symmetry axis of the large umbrella group as the reference, verify the distance of L / 2 in the up and down directions along the centerline direction (L is the length of the rectangle), and define the final large umbrella group region as , the module outputs all identified large umbrella group regions on the composite insulator , effectively distinguishing real heat from reflection interference.

[0035] The module effectively distinguishes between real and pseudo-heating by using geometric and temperature distribution features, and completes the identification of reflections from large umbrella groups under different lighting conditions.

[0036] (5) The composite insulator central axis reconstruction module removes all the large umbrella group rectangular areas identified in the previous module from the original insulator segmentation mask, forming an intermediate mask containing "holes". Subsequently, image restoration algorithms such as morphological closing operation are used to smoothly interpolate the missing areas, reconstructing a complete insulator outline that is approximately without large umbrella groups. On this basis, skeletonization algorithms (such as the central axis transformation method) are used to extract the topological skeleton of the restored outline as the central axis of the insulator, and the corresponding temperature sequence is extracted from the temperature field matrix according to the coordinates of this axis.

[0037] After excluding the area affected by reflections from the large umbrella group, the composite insulator center axis reconstruction module reconstructs the topological center axis that reflects the geometric characteristics of the main insulator structure through smooth interpolation. The process is as follows: (51) Perform interference area removal for large umbrella clusters, and remove all areas output by the large umbrella cluster reflection area identification module. The large umbrella group region is removed from the original binary segmentation mask matrix M of the composite insulator. This step sets the pixel values ​​of the corresponding large umbrella group region in the mask from 1 to 0, thus forming a segmentation mask matrix M with the large umbrella group region removed. holey That is, the mask M with holes. holey .

[0038] (52) Smoothly interpolate the holes based on morphological closing operations (such as using circular structural elements for expansion and erosion), and utilize the temperature constraint of the insulator body region (such as using the average temperature of the centerline) (As a filling reference), temperature-constrained morphological restoration is performed to obtain the segmentation mask matrix M after removing the large umbrella cluster. holey The average temperature of the pixel located at the center line of the entire composite insulator is used. As the filler value, the final result is M after repairing and removing the large umbrella cluster. repaired .

[0039] (53) Obtain the complete outline M after removing and repairing the large umbrella cluster. repaired Then, the pixel coordinates of the geometric center line of the minimum bounding rectangle of the pre-defined composite insulator are reused to extract the corresponding temperature sequence from the temperature matrix after the removal of the large umbrella group and the repair and filling in step (52). This temperature sequence is used to determine whether the corresponding composite insulator is heating up by a trained one-dimensional convolutional neural network. That is, the skeletonization algorithm (such as the median transformation method) is applied to extract M. repairedthe topological skeleton as the central axis of reconstruction, and extracts temperature sequences from the temperature matrix according to the axis coordinates.

[0040] (6) The heating defect intelligent diagnosis and interference suppression module judges heating through a well-trained one-dimensional convolutional neural network (1D-CNN) on the central axis temperature sequence.

[0041] The network model is pre-trained using a large number of labeled heating and normal samples, and can learn the weak abnormal patterns in the temperature sequence. After inputting the temperature sequence, the network extracts local features through the convolutional layer, reduces the dimension through the pooling layer, and outputs the heating probability through the fully connected layer. Combined with the threshold value judgment, the final heating defect diagnosis result is output, and the false positives caused by the reflection of large umbrella groups are automatically suppressed, realizing interference suppression.

[0042] Through experiments, it is verified that the present application converts the complex infrared image analysis into fine analysis of temperature field and geometric characteristics through multi-module collaborative processing, effectively overcomes the problem of large umbrella group reflection interference. By introducing the strategy of combining deep learning and morphology, not only the accuracy of target detection and segmentation is improved, but also the robustness of central axis reconstruction and interference suppression is ensured. The experimental results show that the method can significantly reduce the false positive rate and improve the detection accuracy of real heating defects. The results are highly consistent with the artificial expert interpretation results, and the processing efficiency is high, which is suitable for large-scale unmanned aerial vehicle inspection scenes. The present application provides reliable technical support for intelligent operation and maintenance of power transmission line composite insulators, and effectively improves the automation level of safe operation of power systems.

Claims

1. A method for judging and suppressing interference in the heating defect of a large umbrella group of composite insulators, characterized in that, The module includes an infrared image acquisition and preprocessing module, a composite insulator target detection and positioning module, a composite insulator temperature field analysis module, a large umbrella group reflection area identification module, a composite insulator central axis reconstruction module, and a heating defect intelligent diagnosis and interference suppression module. The specific processing flow is as follows: (1) The infrared image acquisition and preprocessing module interacts with the UAV inspection platform through the preset OBS communication interface to acquire infrared images. At the same time, the preprocessing process includes strict access verification of the input images, filtering based on pixel resolution, and removing image data that does not meet the format requirements; and verifying the source of the image metadata by parsing it to ensure the standardization of the input data. (2) The composite insulator target detection and localization module uses an image recognition algorithm and a deep learning-based instance segmentation algorithm to analyze the input image. The detection head and the segmentation head work together to output the result and generate the minimum bounding rectangle of the composite insulator. (3) The composite insulator temperature field analysis module uses the pixel coordinates of the segmented region of the insulator determined by the composite insulator target detection and positioning module as an index to extract the temperature at the corresponding position from the global radiation temperature matrix returned by the SDK and form the corresponding temperature matrix. (4) The large umbrella cluster reflection area identification module identifies the "large umbrella cluster" structure with a size significantly larger than the standard umbrella skirt by performing morphological analysis on the segmented area of ​​the insulator. Then, it analyzes the temperature distribution characteristics of these areas: it focuses on identifying those areas that exhibit uniform high brightness, have a gentle temperature gradient, and form a sudden change in temperature with the surrounding umbrella skirt. Finally, it positions each identified large umbrella cluster area as a compact rectangular area and records its coordinates to provide spatial positioning for subsequent interference elimination. (5) The composite insulator central axis reconstruction module removes all the large umbrella group rectangular areas identified by the previous module from the original insulator segmentation mask to form an intermediate mask containing "holes". Then, the image repair algorithm is used to smoothly interpolate the missing areas to reconstruct a complete insulator outline that is approximately without large umbrella groups. On this basis, the skeletonization algorithm is used to extract the topological skeleton of the repaired outline as the central axis of the insulator, and the corresponding temperature sequence is extracted from the temperature field matrix according to the coordinates of the axis. (6) The intelligent diagnosis and interference suppression module for heating defects uses a trained one-dimensional convolutional neural network to determine the heating of the central axis temperature sequence.

2. The method for judging and suppressing interference of heating defects in large umbrella groups of composite insulators according to claim 1, characterized in that, in In step (1), the infrared image acquisition and preprocessing module obtains raw data from a distributed data source through a standardized pipeline. The module interacts with the UAV inspection platform through a preset OBS communication interface to achieve batch acquisition of infrared images. This module executes a standardized preprocessing and quality assessment pipeline. First, it performs a rigorous entry verification on the input images: a primary screening is conducted based on a fixed pixel resolution to automatically remove image data that does not meet the format requirements; then, by parsing the image file metadata, it verifies the source of the acquisition device, thereby ensuring the standardization of the input data and the reliability of the subsequent temperature analysis process.

3. The method for judging and suppressing interference of heating defects in large umbrella groups of composite insulators according to claim 1, characterized in that, in In step (2), the composite insulator target detection and localization module utilizes a well-trained deep neural network to perform multi-scale feature extraction and fusion based on the backbone network and the feature pyramid network. During the detection process, candidate target regions are first generated using a region extraction protocol and its anchor box mechanism. Then, a classification network determines whether a region contains a composite insulator and regresses its precise bounding box coordinates. Subsequently, a further segmentation head performs pixel-level classification on each confirmed positive candidate box, generating a high-precision binary segmentation mask. This segmentation mask is represented by matrix M, where M(x,y)=1 indicates that pixel (x,y) belongs to the composite insulator target, and M(x,y)=0 indicates that it belongs to the background. By applying a non-maximum suppression algorithm, redundant and overlapping detection results are eliminated based on bounding box confidence and segmentation overlap. Finally, the composite insulator detected in each image is output. This composite insulator is the smallest rotational bounding rectangle that can tightly enclose a tilted or arbitrarily oriented composite insulator, precisely described by a quintuple. in, The coordinates of the center point of the minimum rotational bounding rectangle in the image pixel coordinate system are given; w represents the length of the composite insulator rectangle in its own coordinate system; and h represents the width of the composite insulator rectangle in its own coordinate system. This indicates the rotation angle of the rectangular frame, which is the tilt angle of the composite insulator.

4. The method for judging and suppressing interference of heating defects in large umbrella groups of composite insulators according to claim 1, characterized in that, in In step (3), the composite insulator temperature field analysis module extracts the absolute temperature value corresponding to the pixel region of the composite insulator from the original infrared data based on the segmentation mask output by the composite insulator target detection and localization module, and constructs its dedicated local temperature matrix. Through deep integration and calling of the SDK, the global radiation temperature matrix covering the entire infrared image field of view is obtained through the temperature interface. Where R represents the infrared image pixel matrix, H and W represent the image height and width, respectively, and the global temperature matrix... Each element (i,j) stores the unprocessed raw temperature value corresponding to pixel coordinates (i,j) in the image.

5. The method for judging and suppressing interference of heating defects in large umbrella groups of composite insulators according to claim 4, characterized in that, in In step (3), specifically, the composite insulator temperature field analysis module uses the binary segmentation mask matrix generated by the composite insulator target detection and positioning module. As a spatial index, pixels with a value of 1 in this matrix constitute the precise outline of the composite insulator, obtained through pixel-by-pixel mapping from the global temperature matrix. Extract the temperature values ​​that belong only to the minimum rotational circumscribed matrix, and construct a local temperature matrix from these temperature values. , 。 6. The method for judging and suppressing interference of heating defects in large umbrella groups of composite insulators according to claim 1, characterized in that, in In step (4), the large umbrella group reflection area identification module achieves accurate identification based on the physical characteristics of the reflection area having aggregate symmetry and uniform temperature distribution. The specific process is as follows: (41) Based on the parameters of the minimum circumscribed rectangle OBB of the composite insulator obtained from the composite insulator temperature field analysis module, extract its geometric center line parallel to the length; let the center point of OBB be... The rotation angle is If the length is L and the width is W, then the centerline can be represented by the equation: By sampling along the center line at pixel intervals, an ordered set of points is obtained: in, ; t represents from the center point (x c , y c The distance moved by one pixel along the center line direction (cosθ, sinθ); i represents the pixel on the center line; (42) For each center point To perform the relevant operations: First, the direction vector of the straight line passing through this point and perpendicular to the center line is... Next, determine the sampling line segment Li by finding the intersection of the vertical line and the OBB boundary; then extract the temperature values ​​of all pixels on the line segment Li to form a temperature profile Ti(s); calculate the temperature similarity index of this temperature profile, assuming the center point temperature is Tc=Ti(0), if the temperature value of the temperature profile... If this point is close to the center temperature, then the number N of all similar pixels on this temperature profile is counted. same Compare similar points on this temperature profile with all other points to determine the similarity. If the number of similar points exceeds two-thirds, that is... If N is a vertical line, then this vertical line is considered a candidate reflection line; where N is a vertical line. total This indicates the total number of pixels on the temperature profile. (43) After completing the scanning of all centerlines, perform continuity analysis; let the set of continuous candidate reflection line indices be {i, i+1, ..., i+k}, if the continuous length If k+1, then this area is determined to be a candidate reflection area; (44) In order to accurately locate the reflection area, find the maximum similarity among the continuous candidate lines: The corresponding vertical line Lmax was determined as the axis of symmetry of the large umbrella group; (45) Using the vertical line Lmax, the axis of symmetry of the large umbrella group, as a reference, verify the distance L / 2 both above and below along the center line, and define the final large umbrella group area as... The module outputs all identified large umbrella-shaped areas on this composite insulator. .

7. The method for judging and suppressing interference of heating defects in large umbrella groups of composite insulators according to claim 6, characterized in that, In step (5), after excluding the large umbrella group reflection interference area, the composite insulator center axis reconstruction module reconstructs the topological center axis reflecting the geometric characteristics of the main insulator structure through smooth interpolation. The process is as follows: (51) Perform interference area removal for large umbrella clusters, and remove all areas output by the large umbrella cluster reflection area identification module. The large umbrella group region is removed from the original binary segmentation mask matrix M of the composite insulator. This step sets the pixel values ​​of the corresponding large umbrella group region in the mask from 1 to 0, thus forming a segmentation mask matrix M with the large umbrella group region removed. holey ; (52) Perform temperature-constrained morphological restoration to obtain the segmentation mask matrix M for removing the large umbrella group. holey The average temperature of the pixel located at the center line of the entire composite insulator is used. As the filler value, the final result is M after repairing and removing the large umbrella cluster. repaired ; (53) Obtain the complete outline M after removing and repairing the large umbrella cluster. repaired Then, the pixel coordinates of the geometric center line of the minimum bounding rectangle of the composite insulator are reused to extract the corresponding temperature sequence from the temperature matrix after the removal of the large umbrella group and the repair filling in step (52). This temperature sequence is used to determine whether the corresponding composite insulator is heating up by a trained one-dimensional convolutional neural network.

8. The method for judging and suppressing interference of heating defects in large umbrella groups of composite insulators according to claim 1, characterized in that, in In step (6), the one-dimensional convolutional neural network is pre-trained using labeled heated and normal samples to learn the weak abnormal patterns in the temperature sequence; after inputting the temperature sequence, the network extracts local features through the convolutional layer, reduces the dimension through the pooling layer, and outputs the heating probability through the fully connected layer; combined with threshold judgment, the final output of the heating defect diagnosis result is given, and the false alarms caused by the reflection of the large umbrella group are automatically suppressed to achieve interference suppression.

Images