Electric power inspection method

Abstract

The invention discloses an electric power inspection method. The method comprises the steps of forming an insulator image set; an original insulator detection model is constructed, and training, evaluation and testing are carried out; and acquiring an image of an insulator to be detected, and inputting the image into the insulator detection model for insulator identification. The method has the beneficial effects that the insulator with the fault can be quickly inspected, the cost is low, the efficiency is high, and the safety and the technicality are relatively good. Through fusion and application of a single-stage target detection technology, an acceleration robust feature technology and an anomaly detection technology, the electric power inspection detection accuracy is greatly improved. In the high-voltage power transmission line, the insulator with the fault can be quickly and accurately searched, and the existing suspicious fault point can be efficiently and reasonably inspected, so that the stable operation of the high-voltage power transmission line is guaranteed. The unmanned aerial vehicle can hover above the insulator of the power transmission line in an intelligent and spot-hovering manner for local detection, so that the task load and time of manual inspection are reduced.

Description

Technical Field

[0001] The invention belongs to the technical field of power facility maintenance, and in particular relates to a power inspection method. Background Art

[0002] At present, power inspection mainly relies on two methods: manual inspection and drone inspection. Manual inspection requires high professionalism of inspectors, and the inspection is carried out in complex terrain in the field, which has low inspection efficiency and takes a long time, so it is difficult to ensure the timeliness of the inspection. Drone inspection currently still mostly relies on manual screening of videos and images to find problems, and some rely on analysis algorithms for identification. Traditional analysis algorithms are developed based on YoloV4, but they cannot identify aging of power cables and damaged insulators.

[0003] Chinese Patent Application No. 202310138932.1 discloses "an intelligent power inspection robot", which includes a body, on which a probe is provided, and the probe is composed of a visible light camera and an infrared thermal imager. This invention is a small power inspection robot, which can only inspect ground power equipment and cannot perform inspections in low altitude, high altitude and other areas. This equipment cannot be fully automated and requires human operation. The operation is complicated and cannot achieve the method of automated inspection at all. Although this invention belongs to the field of power inspection technology, it is far from achieving the effect of power inspection.

[0004] Chinese patent application No. CN202310093467.4 discloses a "transmission line bolt missing fault detection method based on improved YOLOX", which aims to overcome the problems of low detection accuracy and slow detection speed of existing detection methods, including: S1, constructing a transmission line missing bolt dataset, and dividing the dataset into a training set, a validation set, and a test set; S2, improving the YOLOX model: replacing the enhanced feature extraction network with a simplified BiFPN module, and adding a CBAM attention mechanism module after each branch; S3, training the improved YOLOX network with the training set to obtain a training model; S4, according to the training model, input the bolt image to be detected, output the detection image, and complete the detection of bolt missing in the transmission line image. This invention only detects bolts in power inspections. The detection of bolts alone requires a large amount of data sets as support. If it is in a new environment, the detection result will be very low. Moreover, the pictures taken by monitoring screws still rely on personnel to inspect, and the purpose of automation has not been achieved. This invention is almost negligible in practical applications. Summary of the invention

[0005] In order to solve the above problems, an object of the present invention is to provide a method for power inspection.

[0006] In order to achieve the above object, the power inspection method provided by the present invention comprises the following steps performed in sequence:

[0007] 1) Use the image acquisition equipment on the drone to photograph, preprocess and annotate the insulators within the power inspection range to form an insulator image set, and then divide the insulator image set into a training set, a validation set and a test set in proportion;

[0008] 2) Construct an original insulator detection model consisting of a deep learning model, a single target detection model, an accelerated robust model, and an anomaly detection model;

[0009] 3) Input the above training set into the original insulator detection model for training, and then use the validation set and test set to evaluate and test the trained insulator detection model, obtain the insulator detection model and deploy it on the UAV power inspection platform;

[0010] 4) Control the UAV to fly along the power inspection route, and use the onboard image acquisition equipment to collect the image of the insulator to be inspected in real time. Then, after preprocessing, input the image into the above-mentioned insulator detection model for insulator recognition, image alignment and anomaly detection. Finally, the state of the insulator is determined. If it is determined to be an insulator in a normal state, its image is used to expand the insulator image set. If it is determined to be an insulator in an abnormal state, the recognition result is given and a detection report is generated.

[0011] In step 1), the specific steps of using the image acquisition device on the drone to photograph, preprocess and annotate the insulators within the power inspection range to form an insulator image set, and then dividing the insulator image set into a training set, a validation set and a test set in proportion are as follows:

[0012] 1.1) Control the drone to fly within the power inspection range, and use the onboard image acquisition equipment to take pictures of various insulators to obtain the original insulator images;

[0013] 1.2) Preprocessing the original insulator image including denoising, adjusting brightness and contrast, cropping and scaling to improve the quality and consistency of the image, and then marking the information including the position and category of the insulator on the preprocessed insulator image;

[0014] 1.3) All labeled insulator images are used as template images to form an insulator image set, and then the insulator image set is divided into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0015] In step 4), the drone is controlled to fly along the power inspection route, and the image of the insulator to be detected is collected in real time by using the airborne image acquisition device, and then input into the above-mentioned insulator detection model after preprocessing to perform insulator recognition, image alignment and abnormality detection. Finally, the state of the insulator is determined. If it is determined to be an insulator in a normal state, its image is used to expand the insulator image set. If it is determined to be an insulator in an abnormal state, the recognition result is given. The specific steps of generating the detection report are as follows:

[0016] 4.1) Control the drone to fly along the power inspection route, and use the onboard image acquisition equipment to collect images of the insulators to be inspected in real time;

[0017] 4.2) performing preprocessing including scaling, cropping and normalization on the above-mentioned insulator image to be detected, so as to facilitate subsequent target detection processing;

[0018] 4.3) Input the preprocessed insulator image to be detected into the above-mentioned insulator detection model, and use the deep learning model yolov5 to extract features of the preprocessed insulator image to be detected to obtain a feature map;

[0019] 4.4) Use the single target detection model YOLO or SSD to detect insulators on the feature map. During the insulator detection process, it is necessary to classify and regress the insulators at each position in the feature map to obtain the category and position information of the insulator at each position, and then perform post-processing to eliminate repeated detections and excessive detection results;

[0020] 4.5) Mark the location and category of the insulator in the corresponding image of the insulator to be inspected and visualize it for easy viewing and analysis by the user;

[0021] 4.6) Using the accelerated robust model SURF, feature extraction is performed on the template image and the image of the insulator to be detected to obtain feature points and feature descriptors of the two images;

[0022] 4.7) Match the feature points of the two images to obtain matching points, and then use the feature descriptor to calculate the distance and similarity between the matching points, and regard two matching points with close similarity as a matching point pair;

[0023] 4.8) Using the matching point pairs, the least square method or RANSAC algorithm is used to calculate the transformation matrix, and then the template image is aligned with the insulator image to be inspected according to the transformation matrix to facilitate subsequent image processing and analysis;

[0024] 4.9) Using the anomaly detection model PaDiM, the feature vectors of each template image and the insulator image to be detected are extracted and aggregated to obtain the global feature vector of each image;

[0025] 4.10) For each insulator image to be detected, the pixel-level anomaly score is calculated using its global feature vector and the global feature vectors of all template images;

[0026] 4.11) Using a preset threshold, the state of the image is determined according to the pixel-level anomaly score of the insulator image to be detected. If the pixel-level anomaly score is lower than the threshold, the insulator in the insulator image to be detected is determined to be in a normal state, and the insulator image to be detected is used to expand the insulator image set; if the pixel-level anomaly score exceeds the threshold, the insulator in the insulator image to be detected is determined to be in an abnormal state, the recognition result is given, and a test report is generated for the staff to make a secondary judgment.

[0027] The power inspection method provided by the present invention has the following beneficial effects:

[0028] 1. Compared with the traditional method of using drones to shoot videos and images for manual screening to find problems, this method can quickly inspect faulty insulators. It is low-cost, high-efficiency, has good safety and technicality, and can improve inspection efficiency by 40%.

[0029] 2. This method greatly improves the accuracy of power inspection by integrating single-stage target detection technology, accelerated robust feature technology and anomaly detection technology, with an accuracy rate of up to 90%.

[0030] 3. In high-voltage transmission lines, faulty insulators can be found quickly and accurately, and the suspected fault points can be inspected efficiently and reasonably, which is the guarantee for the stable operation of high-voltage transmission lines.

[0031] 4. Drones can intelligently hover above the insulators of transmission lines for local inspection, thereby reducing the workload and time of manual inspections. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A flow chart of the method for power inspection provided by the present invention. DETAILED DESCRIPTION

[0033] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] like Figure 1 As shown, the power inspection method provided by the present invention includes the following steps performed in sequence:

[0035] 1) Use the image acquisition equipment on the drone to photograph, preprocess and annotate the insulators within the power inspection range to form an insulator image set, and then divide the insulator image set into a training set, a validation set and a test set in proportion;

[0036] The specific steps are as follows:

[0037] 1.1) Control the drone to fly within the power inspection range, and use the onboard image acquisition equipment to take pictures of various insulators to obtain the original insulator images; during the acquisition process, the influence of factors such as light, angle, and distance on the image quality should be considered to ensure that the image has good clarity and accuracy;

[0038] 1.2) Preprocessing the original insulator image including denoising, adjusting brightness and contrast, cropping and scaling to improve the quality and consistency of the image, and then marking the information including the position and category of the insulator on the preprocessed insulator image;

[0039] 1.3) All labeled insulator images are used as template images to form an insulator image set, and then the insulator image set is divided into a training set, a validation set, and a test set in a ratio of 7:2:1.

[0040] 2) Construct an original insulator detection model consisting of a deep learning model, a single target detection model, an accelerated robust model, and an anomaly detection model;

[0041] 3) Input the above training set into the original insulator detection model for training, and then use the validation set and test set to evaluate and test the trained insulator detection model, obtain the insulator detection model and deploy it on the UAV power inspection platform;

[0042] During the training process, the original insulator detection model needs to be adjusted and optimized to achieve the best detection effect. Evaluation and testing are used to determine the accuracy and robustness of the model. If the model performs poorly, the model needs to be further optimized and improved.

[0043] 4) Control the UAV to fly along the power inspection route, and use the onboard image acquisition equipment to collect the image of the insulator to be inspected in real time. Then, after preprocessing, input the image into the above-mentioned insulator detection model for insulator recognition, image alignment and anomaly detection. Finally, the state of the insulator is determined. If it is determined to be an insulator in a normal state, its image is used to expand the insulator image set. If it is determined to be an insulator in an abnormal state, the recognition result is given and a detection report is generated.

[0044] The specific steps are as follows:

[0045] 4.1) Control the drone to fly along the power inspection route, and use the onboard image acquisition equipment to collect images of the insulators to be inspected in real time;

[0046] 4.2) performing preprocessing including scaling, cropping and normalization on the above-mentioned insulator image to be detected, so as to facilitate subsequent target detection processing;

[0047] 4.3) Input the preprocessed insulator image to be detected into the above-mentioned insulator detection model, and use the deep learning model yolov5 to extract features of the preprocessed insulator image to be detected to obtain a feature map;

[0048] 4.4) Use the single target detection model YOLO or SSD to detect insulators on the feature map. During the insulator detection process, it is necessary to classify and regress the insulators at each position in the feature map to obtain the category and position information of the insulator at each position, and then perform post-processing to eliminate repeated detections and excessive detection results;

[0049] 4.5) Mark the location and category of the insulator in the corresponding image of the insulator to be inspected and visualize it for easy viewing and analysis by the user;

[0050] 4.6) Using the accelerated robust model SURF, feature extraction is performed on the template image and the image of the insulator to be detected to obtain feature points and feature descriptors of the two images;

[0051] 4.7) Match the feature points of the two images to obtain matching points, and then use the feature descriptor to calculate the distance and similarity between the matching points, and regard two matching points with close similarity as a matching point pair;

[0052] 4.8) Using the matching point pairs, the least square method or RANSAC algorithm is used to calculate the transformation matrix, and then the template image is aligned with the insulator image to be inspected according to the transformation matrix to facilitate subsequent image processing and analysis;

[0053] 4.9) Using the anomaly detection model PaDiM, the feature vectors of each template image and the insulator image to be detected are extracted and aggregated to obtain the global feature vector of each image;

[0054] 4.10) For each insulator image to be detected, the pixel-level anomaly score is calculated using its global feature vector and the global feature vectors of all template images;

[0055] 4.11) Using a preset threshold, the state of the image is determined according to the pixel-level anomaly score of the insulator image to be detected. If the pixel-level anomaly score is lower than the threshold, the insulator in the insulator image to be detected is determined to be in a normal state, and the insulator image to be detected is used to expand the insulator image set; if the pixel-level anomaly score exceeds the threshold, the insulator in the insulator image to be detected is determined to be in an abnormal state, the recognition result is given, and a test report is generated for the staff to make a secondary judgment.

Claims

1. A method for power inspection, characterized in that: The power inspection method comprises the following steps performed in sequence: 1) Use the image acquisition equipment on the drone to photograph, preprocess and annotate the insulators within the power inspection range to form an insulator image set, and then divide the insulator image set into a training set, a validation set and a test set in proportion; 2) Construct an original insulator detection model consisting of a deep learning model, a single target detection model, an accelerated robust model, and an anomaly detection model; 3) Input the above training set into the original insulator detection model for training, and then use the validation set and test set to evaluate and test the trained insulator detection model, obtain the insulator detection model and deploy it on the UAV power inspection platform; 4) Control the UAV to fly along the power inspection route, and use the onboard image acquisition equipment to collect the image of the insulator to be inspected in real time. Then, after preprocessing, input the image into the above-mentioned insulator detection model for insulator recognition, image alignment and anomaly detection. Finally, the state of the insulator is determined. If it is determined to be an insulator in a normal state, its image is used to expand the insulator image set. If it is determined to be an insulator in an abnormal state, the recognition result is given and a detection report is generated.

2. The method for power inspection according to claim 1, characterized in that: In step 1), the specific steps of using the image acquisition device on the drone to photograph, preprocess and annotate the insulators within the power inspection range to form an insulator image set, and then dividing the insulator image set into a training set, a validation set and a test set in proportion are as follows: 1.1) Control the drone to fly within the power inspection range, and use the onboard image acquisition equipment to take pictures of various insulators to obtain the original insulator images; 1.2) Preprocessing the original insulator image including denoising, adjusting brightness and contrast, cropping and scaling to improve the quality and consistency of the image, and then marking the information including the position and category of the insulator on the preprocessed insulator image; 1.3) All labeled insulator images are used as template images to form an insulator image set, and then the insulator image set is divided into a training set, a validation set, and a test set in a ratio of 7:2:

1.

3. The method for power inspection according to claim 2, characterized in that: In step 4), the drone is controlled to fly along the power inspection route, and the image of the insulator to be detected is collected in real time by using the airborne image acquisition device, and then input into the above-mentioned insulator detection model after preprocessing to perform insulator recognition, image alignment and abnormality detection. Finally, the state of the insulator is determined. If it is determined to be an insulator in a normal state, its image is used to expand the insulator image set. If it is determined to be an insulator in an abnormal state, the recognition result is given. The specific steps of generating the detection report are as follows: 4.1) Control the drone to fly along the power inspection route, and use the onboard image acquisition equipment to collect images of the insulators to be inspected in real time; 4.2) performing preprocessing including scaling, cropping and normalization on the above-mentioned insulator image to be detected, so as to facilitate subsequent target detection processing; 4.3) Input the preprocessed insulator image to be detected into the above-mentioned insulator detection model, and use the deep learning model yolov5 to extract features of the preprocessed insulator image to be detected to obtain a feature map; 4.4) Use the single target detection model YOLO or SSD to detect insulators on the feature map. During the insulator detection process, it is necessary to classify and regress the insulators at each position in the feature map to obtain the category and position information of the insulator at each position, and then perform post-processing to eliminate repeated detections and excessive detection results; 4.5) Mark the location and category of the insulator in the corresponding image of the insulator to be inspected and visualize it for easy viewing and analysis by users; 4.6) Using the accelerated robust model SURF, feature extraction is performed on the template image and the image of the insulator to be detected to obtain feature points and feature descriptors of the two images; 4.7) Match the feature points of the two images to obtain matching points, and then use the feature descriptor to calculate the distance and similarity between the matching points, and regard two matching points with close similarity as a matching point pair; 4.8) Using the matching point pairs, the least square method or RANSAC algorithm is used to calculate the transformation matrix, and then the template image is aligned with the insulator image to be inspected according to the transformation matrix to facilitate subsequent image processing and analysis; 4.9) Using the anomaly detection model PaDiM, the feature vectors of each template image and the insulator image to be detected are extracted and aggregated to obtain the global feature vector of each image; 4.10) For each insulator image to be detected, the pixel-level anomaly score is calculated using its global feature vector and the global feature vectors of all template images; 4.11) Using a preset threshold, the state of the image is determined according to the pixel-level anomaly score of the insulator image to be detected. If the pixel-level anomaly score is lower than the threshold, the insulator in the insulator image to be detected is determined to be in a normal state, and the insulator image to be detected is used to expand the insulator image set; if the pixel-level anomaly score exceeds the threshold, the insulator in the insulator image to be detected is determined to be in an abnormal state, the recognition result is given, and a test report is generated for the staff to make a secondary judgment.

Images