Infrared temperature anomaly detection method for insulator strings of power equipment based on instance segmentation

Through DDE technology enhancement and instance segmentation of YOLACT model, combined with the isolated forest algorithm, the inefficiency and false detection problems in the temperature abnormality detection of power equipment insulator strings are solved, and higher accuracy detection is achieved, reducing operation and maintenance costs.

CN118470051BActive Publication Date: 2024-11-12STATE GRID JIANGXI ELECTRIC POWER CO LTD RES INST +3
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410932422.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-07-12
Publication Date
2024-11-12
Estimated Expiration
2044-07-12

AI Technical Summary

Technical Problem

When detecting temperature abnormalities of insulator strings of power equipment, the prior art is inefficient and prone to false detection due to complex backgrounds, and it is impossible to effectively identify the temperature abnormalities of insulator strings in complex power image data.

Method used

An instance-based segmentation method is adopted to enhance infrared images through DDE technology, and the real-time segmentation of insulator strings is used using the YOLACT model, combined with the isolated forest algorithm, reduce background noise interference and improve detection accuracy.

Benefits of technology

It improves the accuracy of temperature abnormality detection of insulator strings, reduces the cost of power grid operation and maintenance, saves manpower and time, and enhances detection efficiency and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118470051B_ABST
    Figure CN118470051B_ABST
Patent Text Reader

Abstract

The present invention discloses an infrared temperature anomaly detection method for an insulator string of an electric power equipment based on instance segmentation. The infrared image of the insulator string is enhanced by DDE technology, and then segmented by using YOLACT instance segmentation model to obtain an insulator string region image with background removed; radial temperature distribution characteristic curves of n insulator slice image regions are drawn by infrared temperature measurement technology; the insulator string region image with background removed is trained by using isolation forest algorithm to obtain an isolation forest model for calculating the abnormal score of the insulator slice; the radial temperature distribution characteristic curves of the n insulator slice image regions are fitted to obtain a fitting curve, and the determination coefficient R of the fitting curve is used to calculate the abnormal score of the insulator slice. 2 , according to the coefficient of determination R 2 The invention improves the accuracy of infrared temperature anomaly detection of insulator strings.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the field of infrared detection technology, and in particular relates to an infrared temperature anomaly detection method for an insulator string of a power equipment based on instance segmentation. Background Art

[0002] Insulators are indispensable components in overhead lines. They play a dual role in preventing current from grounding and supporting conductors. However, abnormal heating caused by defects in the insulator interface occurs frequently, which will have a significant impact on the normal operation of power lines. Therefore, temperature anomaly detection of insulator equipment in power lines is essential.

[0003] If only manpower is used to detect temperature anomalies in insulator equipment, it will consume a lot of manpower and material resources and be inefficient. The red-hot external imaging technology has been used to study the heating mechanism and temperature distribution anomalies of insulator strings. The red-hot external imaging technology can only solve the temperature anomaly identification of a single insulator string, but for power image data with complex backgrounds, some backgrounds are very likely to cause false detections. Summary of the invention

[0004] The purpose of the present invention is to improve an infrared temperature anomaly detection method for an insulator string of an electric power equipment based on instance segmentation. The present invention can improve the accuracy of identifying temperature anomaly defects in insulators of power lines; the noise interference of the image background during temperature anomaly detection is avoided through the instance segmentation algorithm, the accuracy is improved, and the operation and maintenance cost of the power grid is reduced.

[0005] The present invention is implemented by the following technical solutions: A method for detecting infrared temperature anomaly of insulator strings of power equipment based on instance segmentation, the steps are as follows:

[0006] Step S1, performing DDE technology enhancement on the collected infrared image of the insulator string;

[0007] Step S2, annotating the infrared image of the insulator string enhanced by the DDE technology, and constructing standard training data of the infrared image of the insulator string;

[0008] Step S3, training the YOLACT model using standard training data of infrared images of insulator strings to determine whether the accuracy is achieved, until a YOLACT instance segmentation model that can accurately segment the insulator strings is obtained;

[0009] Step S4, using the trained YOLACT instance segmentation model to segment the real-time insulator string infrared image enhanced by the DDE technology, outputting the insulator string segmentation result, and obtaining the insulator string region image with the background removed;

[0010] Step S5, dividing the insulator string region image into n insulator segment image regions according to the shape feature, traversing the pixels of the insulator segment image region to obtain the temperature by comparing with the grayscale of the legend, and then calculating the relative temperature values ​​of the n insulator segment image regions in turn; and drawing radial temperature distribution characteristic curves of the n insulator segment image regions by infrared temperature measurement technology;

[0011] Step S6, using the insulator string region image with the background removed to train the isolation forest algorithm, using the known features of the normal and abnormal insulator slice image regions to train the isolation forest algorithm, to obtain an isolation forest model capable of identifying abnormal insulator slices, which is used to calculate the abnormality score of the insulator slice; calculating the abnormality score of the insulator slice of the insulator string region image with the background removed obtained in step S4 by using the isolation forest model;

[0012] Step S7: Fitting the radial temperature distribution characteristic curves of the n insulator slice image regions to obtain a fitting curve, and using the determination coefficient R of the fitting curve 2 , according to the coefficient of determination R 2 The abnormal score of the insulator segment is used to comprehensively judge whether the insulator string has temperature anomaly.

[0013] Specifically, when R 2 ≥0.6 and the abnormal score of the insulator piece is lower than the set threshold, it is judged that there is no temperature abnormality; when R 2 When the value is less than 0.6 or the abnormal score of the insulator segment is higher than the set threshold, it is determined that there is a temperature abnormality.

[0014] Specifically, the labelme annotation tool is used to annotate the infrared image of the insulator string enhanced by the DDE technology.

[0015] Specifically, the DDE technology enhancement process is as follows: first, the infrared image of the insulator string is separated into low-frequency components and high-frequency components using a low-pass filter, and the low-frequency components are subjected to histogram processing, such as histogram partitioning and histogram equalization; then, the results of the low-frequency components and high-frequency components subjected to the histogram processing are subjected to linear weighted compression to obtain an enhanced infrared image of the insulator string.

[0016] Specifically, the YOLACT instance segmentation model performs the following segmentation process on the real-time infrared image of the insulator string enhanced by the DDE technology:

[0017] Step S41, inputting the infrared image of the insulator string to be segmented after being enhanced by the DDE technology;

[0018] Step S42, extracting features of the insulator string from the infrared image of the insulator string enhanced by the DDE technology through a feature extraction network to obtain feature maps of different sizes;

[0019] Step S43, fusing feature maps of different sizes through a feature fusion network;

[0020] Step S44, learning the features of the target instance in the feature map through the mask prototype generation branch, and generating semantic information and shape information containing the target;

[0021] Step S45, predicting the mask coefficient for each candidate box through the mask coefficient prediction branch and using the Fast NMS algorithm to obtain the position of the instance in the feature map;

[0022] Step S46, use matrix multiplication fusion to obtain the prediction frame, shorten the model reasoning time, and finally use the segmentation threshold to complete the segmentation of the insulator string.

[0023] Specifically, the feature extraction network adopts the Swin-Transformer backbone network.

[0024] Specifically, the feature fusion network adopts the Feature Pyramid Network (FPN).

[0025] Advantages of the present invention:

[0026] (1) The use of DDE technology enhances the contrast and detail information of infrared images, making the outline and local features of the insulator string clearer, which is conducive to subsequent temperature anomaly detection.

[0027] (2) YOLACT real-time segmentation is performed based on DDE technology, avoiding the manual labeling process in traditional methods, saving time and manpower and reducing costs.

[0028] (3) The method of regional division is used for the segmented insulator string, which can more intuitively examine the temperature distribution of each area of ​​the insulator string and facilitate the detection and location of abnormal areas.

[0029] (4) The segmentation algorithm used for the insulator string can effectively reduce the interference information brought by the image background and improve the accuracy of temperature anomaly detection.

[0030] (5) The isolation forest algorithm is highly efficient. On a large-scale insulator string dataset, its efficiency in detecting abnormal insulator areas is usually better than that of distance- or density-based anomaly detection methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The figure is a flow chart of the method of the present invention.

[0032] Figure 2 Enhanced process flow chart for DDE technology.

[0033] Figure 3 Schematic diagram of the YOLACT instance segmentation model structure. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0035] Reference Figure 1 , a method for detecting infrared temperature anomaly of insulator strings of power equipment based on instance segmentation, the steps are as follows:

[0036] Step S1, performing DDE technology enhancement on the collected infrared image of the insulator string;

[0037] Step S2, using the labelme annotation tool to annotate the infrared image of the insulator string enhanced by the DDE technology, and construct standard training data of the infrared image of the insulator string;

[0038] Step S3, training the YOLACT model using standard training data of infrared images of insulator strings to determine whether the accuracy is achieved, until a YOLACT instance segmentation model that can accurately segment the insulator strings is obtained;

[0039] Step S4, using the trained YOLACT instance segmentation model to segment the real-time insulator string infrared image enhanced by the DDE technology, outputting the insulator string segmentation result, and obtaining the insulator string region image with the background removed;

[0040] Step S5: divide the insulator string region image into n insulator segment image regions according to the shape feature, traverse the pixels of the insulator segment image region to obtain the temperature by comparing with the grayscale of the legend, and then calculate the relative temperature values ​​of the n insulator segment image regions in turn; draw the radial temperature distribution characteristic curves (F1, F2, ..., F n ), F n A characteristic curve of radial temperature distribution in the nth insulator segment image area;

[0041] Step S6, using the insulator string region image with the background removed to train the isolation forest algorithm, using the known features of the normal and abnormal insulator slice image regions to train the isolation forest algorithm, to obtain an isolation forest model capable of identifying abnormal insulator slices, which is used to calculate the abnormality score of the insulator slice; calculating the abnormality score of the insulator slice of the insulator string region image with the background removed obtained in step S4 by using the isolation forest model;

[0042] Step S7: Fitting the radial temperature distribution characteristic curves of the n insulator slice image regions to obtain a fitting curve, and using the determination coefficient R of the fitting curve 2 , according to the coefficient of determination R 2 The abnormal score of the insulator segment is used to comprehensively judge whether the insulator string has temperature anomaly. 2 ≥0.6 (indicates that the fitting curve can explain most of the variation in the data and the fitting effect is good) and the abnormal score of the insulator piece is lower than the set threshold, it is judged that there is no temperature abnormality; when R 2 <0.6(R 2 The closer it is to 0, the more likely it is that the fitting curve cannot explain the variation of the target variable) or when the abnormal score of the insulator piece is higher than the set threshold, it is determined that there is a temperature anomaly.

[0043] The YOLACT instance segmentation model segments the insulator string, avoiding noise interference from the image background during temperature anomaly detection. It can quickly collect temperature data and record the abnormal temperature rise of the insulator in the infrared image of the insulator string. The YOLACT instance segmentation model has good real-time performance and can be used in actual line patrol work. In addition, the isolation forest algorithm performs well in processing high-dimensional data and anomaly detection, which can greatly improve the detection effect of abnormal insulator areas.

[0044] like Figure 2 As shown, the infrared image of the insulator string is first separated into low-frequency components and high-frequency components by low-pass filtering, and the low-frequency components are subjected to histogram processing, such as histogram partitioning and histogram equalization; then the results of the low-frequency components and high-frequency components subjected to histogram processing are linearly weighted compressed to obtain an enhanced infrared image of the insulator string. This embodiment separates the background layer and detail layer of the infrared image of the insulator string by DDE technology, and performs corresponding grayscale enhancement and noise suppression, synthesizes an infrared image of the insulator string by adjusting and compressing the dynamic range of each layer, and uses the temperature difference between the abnormal insulator and other areas to increase the local detail information of the abnormal insulator string in the infrared image of the insulator string; finally, data enhancement of the abnormal part of the insulator is achieved.

[0045] like Figure 3 As shown in the figure, the YOLACT instance segmentation model performs the segmentation process of the real-time insulator string infrared image enhanced by the DDE technology as follows:

[0046] Step S41, inputting the infrared image of the insulator string to be segmented after being enhanced by the DDE technology;

[0047] Step S42, extracting features of the insulator string from the infrared image of the insulator string enhanced by the DDE technology through a feature extraction network to obtain feature maps of different sizes; the feature extraction network adopts a Swin-Transformer backbone network;

[0048] Step S43, fusing feature maps of different sizes through a feature fusion network; the feature fusion network adopts a feature pyramid network (FPN);

[0049] Step S44, learning the features of the target instance in the feature map through the mask prototype generation branch (Protonet), generating semantic information and shape information containing the target;

[0050] Step S45, predicting the mask coefficient for each candidate box through the mask coefficient prediction branch (Prediction head) and using the Fast NMS algorithm to obtain the location of the instance in the feature map;

[0051] Step S46, use matrix multiplication fusion to obtain the prediction frame, shorten the model reasoning time, and finally use the segmentation threshold to complete the segmentation of the insulator string.

[0052] This embodiment adopts the YOLACT network with Swin-Transformer as the backbone network, which effectively improves the segmentation effect of the insulator string area by enhancing the integration of non-local information and the function of extracting global semantic information from high-dimensional features, while significantly reducing the consumption of computing resources.

[0053] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting infrared temperature anomaly of insulator strings of power equipment based on instance segmentation, characterized in that: Here are the steps: Step S1, performing DDE technology enhancement on the collected infrared image of the insulator string; Step S2, annotating the infrared image of the insulator string enhanced by the DDE technology, and constructing standard training data of the infrared image of the insulator string; Step S3, training the YOLACT model using standard training data of infrared images of insulator strings to determine whether the accuracy is achieved, until a YOLACT instance segmentation model that can accurately segment the insulator strings is obtained; Step S4, using the trained YOLACT instance segmentation model to segment the real-time insulator string infrared image enhanced by the DDE technology, outputting the insulator string segmentation result, and obtaining the insulator string region image with the background removed; Step S5, dividing the insulator string region image into n insulator segment image regions according to the shape feature, traversing the pixels of the insulator segment image region to obtain the temperature by comparing with the grayscale of the legend, and then calculating the relative temperature values ​​of the n insulator segment image regions in turn; and drawing radial temperature distribution characteristic curves of the n insulator segment image regions by infrared temperature measurement technology; Step S6, using the insulator string region image with the background removed to train the isolation forest algorithm, using the known features of the normal and abnormal insulator slice image regions to train the isolation forest algorithm, to obtain an isolation forest model capable of identifying abnormal insulator slices, which is used to calculate the abnormality score of the insulator slice; calculating the abnormality score of the insulator slice of the insulator string region image with the background removed obtained in step S4 by using the isolation forest model; Step S7: Fitting the radial temperature distribution characteristic curves of the n insulator slice image regions to obtain a fitting curve, and using the determination coefficient R of the fitting curve 2 , according to the coefficient of determination R 2 Combined with the abnormal score of the insulator segment, it is determined whether the insulator string has temperature anomaly; Among them, the YOLACT instance segmentation model performs the segmentation process of the real-time insulator string infrared image enhanced by DDE technology as follows: Step S41, inputting the infrared image of the insulator string to be segmented after being enhanced by the DDE technology; Step S42, extracting features of the insulator string from the infrared image of the insulator string enhanced by the DDE technology through a feature extraction network to obtain feature maps of different sizes; the feature extraction network adopts a Swin-Transformer backbone network; Step S43, fusing feature maps of different sizes through a feature fusion network; Step S44, learning the features of the target instance in the feature map through the mask prototype generation branch, and generating semantic information and shape information containing the target; Step S45, predicting the mask coefficient for each candidate box through the mask coefficient prediction branch and using the Fast NMS algorithm to obtain the position of the instance in the feature map; Step S46, use matrix multiplication fusion to obtain the prediction frame, shorten the model reasoning time, and finally use the segmentation threshold to complete the segmentation of the insulator string.

2. The method for detecting infrared temperature anomaly of insulator strings of power equipment based on instance segmentation according to claim 1 is characterized in that: The comprehensive conditions for judging whether the insulator string has temperature anomaly are: when R 2 ≥0.6 and the abnormal score of the insulator piece is lower than the set threshold, it is judged that there is no temperature abnormality; when R 2 When the value is less than 0.6 or the abnormal score of the insulator segment is higher than the set threshold, it is determined that there is a temperature abnormality.

3. The method for detecting infrared temperature anomaly of insulator strings of power equipment based on instance segmentation according to claim 1, characterized in that: The labelme annotation tool is used to annotate the infrared image of the insulator string enhanced by DDE technology.

4. The method for detecting infrared temperature anomaly of insulator strings of power equipment based on instance segmentation according to claim 1, characterized in that: The DDE technology enhancement process is as follows: first, the infrared image of the insulator string is separated into low-frequency components and high-frequency components by low-pass filtering, and the low-frequency components are processed by histogram; then, the results of the low-frequency components and high-frequency components that have undergone histogram processing are linearly weighted compressed to obtain an enhanced infrared image of the insulator string.

5. The method for detecting infrared temperature anomaly of insulator strings of power equipment based on instance segmentation according to claim 1, characterized in that: The feature fusion network adopts feature pyramid network.