Power transmission line insulator string identification method based on unmanned aerial vehicle oblique photography

Abstract

The invention provides a power transmission line insulator string identification method based on unmanned aerial vehicle oblique photography. A multi-rotor unmanned aerial vehicle carries a full-frame camera to collect oblique photography image data of a power transmission line and the surrounding environment of the power transmission line, and a YOLO deep learning model is utilized to perform automatic identification of an insulator string. The method aims at improving the automation level and accuracy of routing inspection, reducing the dependence on manpower and improving the operation efficiency. The method specifically comprises the steps of data acquisition, image preprocessing, three-dimensional modeling of the power transmission line and automatic identification of the insulator string. A method of combining deep learning and three-dimensional modeling is utilized to automatically identify an insulator string, and stable operation of a power transmission line is ensured.

Description

Technical Field

[0001] The present invention belongs to the field of oblique photogrammetry three-dimensional modeling and component extraction thereof, and relates to a three-dimensional model reconstruction of a power transmission line based on oblique photogrammetry data, and automatic identification of insulator strings of the power transmission line using a YOLO deep learning method. Background Art

[0002] Insulator strings are important components on transmission lines. They are usually composed of two or more insulator elements and are flexibly suspended on the transmission lines. Insulator strings are protective devices with fixed and operational needs. They are used to suspend conductors and insulate the conductors from the towers and the ground. Therefore, regular monitoring of the safety status of transmission lines and important facility components such as insulator strings is of great significance to the normal and safe operation of the power grid.

[0003] In order to ensure a stable and continuous supply of electricity, the power sector needs to spend a lot of manpower, material and financial resources every year to inspect transmission lines. The traditional inspection method is that line inspectors drive or walk to use instruments or visually inspect power facilities. However, due to the long transmission lines, complex terrain along the lines, lakes, mountains, forests and wastelands, and large climate differences, inspection work is extremely inconvenient, and new technologies are urgently needed to solve this problem. In recent years, remote sensing technology has been fully applied in power inspections with its unique advantage of non-contact data acquisition. Among them, oblique photogrammetry technology is to carry multiple sensors on a flying platform and collect images from five different angles, such as a vertical, four oblique, etc., to reflect the real situation of the ground objects. Based on the characteristics of large-scale mapping of aerial photography, this technology can use the captured optical images for texture mapping to achieve rapid three-dimensional modeling of target objects.

[0004] The existing automatic inspection of transmission lines based on UAV data mostly uses ordinary two-dimensional optical images, while UAV oblique photogrammetry technology can use stereo matching to achieve three-dimensional reconstruction of transmission lines. On the basis of realizing the image acquisition of the transmission line area, it can also realize topographic mapping and geometric measurement of related components. At the same time, the transmission line and its surrounding environment appear more intuitive in the three-dimensional model, and the relative positions of some key components can be accurately grasped, thereby realizing real-time control of the security of the power grid. The three-dimensional modeling steps of oblique photography mainly include extracting image connection points, image dense matching, regional network adjustment solution, aerial triangulation, etc. On the basis of constructing a good three-dimensional model using the original oblique photogrammetry data, a series of automatic methods can be used to identify insulator strings. For example, through digital image processing methods, traditional feature construction methods are used to identify insulator strings based on the feature information of insulator strings in the image, such as line segment characteristics, corner point characteristics, and texture characteristics. However, most of these methods require high-resolution two-dimensional aerial or satellite remote sensing images, and require a prior investigation of the approximate spatial position of the insulator; or they have great limitations in feature extraction, making it difficult to achieve good results. Therefore, this method is based on the three-dimensional model of the transmission line constructed by oblique photogrammetry and uses deep learning methods to automatically identify insulator strings. Summary of the invention

[0005] The purpose of the present invention is to provide a method for automatic identification of transmission line insulator strings based on unmanned aerial vehicle oblique photogrammetry, aiming to improve the automation and accuracy of transmission line inspections, reduce labor costs, and improve work efficiency. Traditional transmission line inspection methods often rely on manual naked eye identification, which is not only inefficient, but also unable to achieve accurate insulator string identification in complex environments. The present invention combines deep learning algorithms with unmanned aerial vehicle oblique photogrammetry to achieve efficient and automatic identification of transmission line insulator strings. By using a multi-rotor unmanned aerial vehicle equipped with a full-frame camera, image data is collected according to a preset route in the study area to obtain oblique photogrammetry image data covering the transmission line. At the same time, aerial photography targets are arranged on the ground to provide a reference standard for geographic location calibration. The collected image data is converted in format and coordinate system to ensure data compatibility and processing consistency. The matching accuracy of the image is checked using ground control points, and the image quality is optimized by image enhancement and denoising technology to facilitate subsequent three-dimensional modeling and identification. Aerial triangulation is performed based on aerial survey software, and a three-dimensional model is reconstructed through the overlapping relationship between images and ground control points, including terrain, towers, power lines and insulator strings. This process combines the principles of multi-view geometry and improves the accuracy of the model through multiple iterative optimizations.

[0006] The technical solution provided by the present invention is: a method for automatically identifying insulator strings of transmission lines based on UAV oblique photogrammetry, the specific process is as shown in the attached Figure 1As shown, it mainly includes the following steps:

[0007] 1) UAV oblique photogrammetry data collection

[0008] 2) UAV image data preprocessing

[0009] 3) Construction of 3D model of transmission lines and surrounding objects

[0010] 4) Automatic identification of insulator strings BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 Flow chart of the method of the present invention.

[0012] Figure 2 Flowchart for obtaining oblique photogrammetric imagery from UAVs.

[0013] Figure 3 It is composed of unmanned aerial vehicle systems.

[0014] Figure 4 This is the location map of the study area.

[0015] Figure 5 A drone route map.

[0016] Figure 6 This is a map of ground control point information collection and its distribution.

[0017] Figure 7 This is the ground control point accuracy analysis.

[0018] Figure 8 This is a transmission line model diagram.

[0019] Fig. 9 This is the YOLO network structure diagram.

[0020] Fig.10 Result diagram for identifying insulator strings. DETAILED DESCRIPTION

[0021] (1) UAV oblique photogrammetry data collection

[0022] The data collection process is as follows Figure 2 As shown in Figure 1, a multi-rotor drone equipped with a full-frame camera is used to collect image data. In this method, the drone equipment is taken as an example of DJI M350 RTK equipped with DJI Zenmuse P1 (see Figure 1). Figure 3 ), the study area is Yushan Town, Kunshan City, Suzhou City, Jiangsu Province (approximately 120.89°E 31.35°N) as an example (Appendix Figure 4). Pre-flight preparations include setting the flight area on Google Earth and saving it as a KML file, which is then imported into the drone system to plan the designated flight area. Before the official flight, by setting basic aerial photography parameters such as the relative photography reference plane line height, circling radius, overlap rate, number of acquisitions, and camera pitch angle, a KML file is generated to achieve human-computer interaction to complete the aerial photography task. When setting parameters, it is necessary to ensure that the relative aerial photography reference plane corresponds to the plane where the transmission line is located. If there is a height difference on the ground, the corresponding compensation height should be set to ensure that the data can be focused near the center of the transmission line during collection, thereby ensuring the precision and accuracy of the three-dimensional model. Use an S-shaped parallel route (attached Figure 5 ) flight mode to collect all-round images of the transmission line and surrounding objects. In order to ensure the accuracy of the data and the authenticity of the model, it is also necessary to deploy aerial photography targets on the ground as ground control points (see Appendix Figure 6 ), using GNSS equipment to accurately measure the plane coordinates and elevation of these control points (see Appendix Figure 7 ). The layout of ground control points should be strictly in accordance with the standards, and locations with wide views, easy preservation and convenient measurement should be selected to ensure the geographic location calibration of the image data.

[0023] (2) UAV image data preprocessing

[0024] After the image data is collected, it needs to be preprocessed to ensure the compatibility and consistency of the data in subsequent processing. The first step in image preprocessing is to convert the data format and coordinate system, and ensure the standardized processing of the image through a unified format and coordinate system. Then, the ground control points are used to check the matching accuracy of the image to ensure the accuracy of the position and direction of each image in the geographic space. During the processing, it is also necessary to screen and classify POS data to eliminate data anomalies caused by external interference during flight. Image enhancement and denoising technology are used to optimize image quality, improve image clarity and detail, and provide high-quality input data for subsequent 3D model construction and deep learning recognition.

[0025] (3) Construction of 3D model of transmission lines and surrounding objects

[0026] Aerial triangulation is performed using aerial survey software, and the spatial coordinates of each pixel are determined by the overlapping relationship between images and known ground control points. This process involves the principles of multi-view geometry, including feature point extraction, matching, and spatial reconstruction. The theoretical basis of aerial triangulation is the collinearity condition equation, and its mathematical expression is:

[0027]

[0028] Among them, x, y are the image plane coordinates of the image point; x0, y0, f are the internal orientation elements of the image; X S ,YS ,Z S X is the object space coordinate of the camera station; A ,Y A ,Z A is the object space coordinate of the object point; a i ,b i ,c i (i=1,2,3) is the object space coordinate of the object point;

[0029] After completing the aerial triangulation calculation, professional modeling software is used to reconstruct the transmission line and its surrounding objects in three dimensions, including the models of terrain, towers, power lines and insulator strings. In order to ensure the accuracy of the model, multiple iterations of optimization are required, and the model is corrected and modified using accuracy evaluation indicators such as smoothness and detail retention to ensure the accuracy and reliability of the reconstructed model. Some models are shown in the attached figure. Figure 8 .

[0030] (4) Automatic identification of insulator strings

[0031] The YOLO deep learning method is used to automatically identify insulator strings. The network structure is shown in the attached figure. Fig. 9 . First, prepare annotated training data sets, which include images of insulator strings taken at multiple angles and under multiple lighting conditions, and manually annotate these images for model training. During the training process, a convolutional neural network model is used for learning. The input is the preprocessed three-dimensional model image, and the output is the identified insulator string position. In order to improve the generalization ability of the model, data enhancement technology is used, including operations such as geometric changes, color space changes, and mixed changes, to increase the diversity of training data. At the same time, cross-validation and hyperparameter tuning are used to further improve the recognition accuracy of the model. After the model training is completed, the trained model is used to automatically identify the insulator strings of the transmission line (see Appendix). Fig.10 ). The identified results need to be evaluated for accuracy to verify the performance of the model. The performance of the model is evaluated using indicators such as accuracy, precision, recall and F1-Score. Accuracy refers to the ratio of the number of samples correctly predicted by the model to the total number of samples. Accuracy is used to evaluate the accuracy of the model in samples predicted as positive examples. Recall is used to evaluate the model's predictive ability for positive examples. The F1 score is the harmonic mean of the precision and recall rates to comprehensively evaluate the performance of the model. The mathematical expressions of accuracy, precision, recall and F1-score are:

[0032]

[0033] Among them, TP (True Positive) means that the predicted result is a positive sample, and the actual sample is also a positive sample, that is, the number of positive samples correctly identified; FP (False Positive) means that the predicted result is a positive sample, but the actual sample is a negative sample, that is, the number of false positive negative samples; TN (True Negative) means that the predicted result is a negative sample, but the actual sample is a positive sample, that is, the number of missed positive samples; FN (True Negative) means that the predicted result is a negative sample, but the actual sample is a negative sample, that is, the number of negative samples correctly identified.

[0034] The results show that the accuracy of this method in insulator string identification is 91.27%, the precision is 92.61%, the recall is 90.21%, and the F1 score is 91.22%. Through the accuracy evaluation, the model can be further improved and optimized to ensure that the final identification result meets the requirements of transmission line safety monitoring.

Claims

1. A method for identifying insulator strings of power transmission lines based on oblique photography by unmanned aerial vehicles, characterized in that: Collect UAV oblique photogrammetry images of the transmission line and its surrounding features, build a 3D model of the transmission line and its surrounding features after preprocessing, and then automatically identify insulator strings in combination with deep learning, including the following steps: (1) UAV oblique photogrammetry data collection (2) UAV image data preprocessing (3) Construction of 3D model of transmission lines and surrounding objects (4) Automatic identification of insulator strings.

2. The method according to claim 1, characterized in that: The UAV oblique photogrammetry data collection includes: using a multi-rotor UAV equipped with a full-frame camera, and adopting an S-shaped parallel flight mode to collect all-round images of the power transmission line and surrounding objects. In order to ensure high-quality data collection, it is necessary to lay out aerial photography target plates on the ground, and use GNSS equipment to accurately measure their plane coordinates and elevation information to verify the accuracy of the data. After the quality of the collected data is evaluated, if there are incomplete or non-compliant areas, supplementary flight design is required to ensure data integrity.

3. The method according to claim 1, characterized in that The preprocessing of the UAV oblique photogrammetry data includes: firstly, converting the data format and the coordinate system to ensure that the data can be used correctly in subsequent processing. Next, the matching accuracy of the image is checked using known ground control points to ensure the accuracy of the image in spatial position and direction. In addition, the original POS data is screened and classified to remove abnormal data to ensure the reliability of the data.

4. The method according to claim 1, characterized in that: The construction of the three-dimensional model of the power transmission line and its surrounding features includes: using aerial survey software to perform aerial triangulation calculations to generate accurate spatial position data. After the aerial triangulation calculations are completed, professional modeling software is used to reconstruct the three-dimensional model, and through multiple iterative optimizations, a high-precision three-dimensional model of the power transmission line and its surrounding features is finally generated.

5. The method according to claim 1, characterized in that The automatic identification of insulator strings comprises: Automatic recognition of insulator strings is achieved based on deep learning models. Specifically, advanced deep learning algorithms are used to extract features and identify targets from image data. In order to evaluate the recognition accuracy, indicators such as accuracy, precision, recall, and F1-Score are used to evaluate the accuracy of the recognition results.

6. The method according to claim 5, characterized in that The recognition process includes: In order to further improve the robustness and accuracy of recognition, the recognition process uses data enhancement operations, including geometric changes, color space changes, and mixed changes, to increase the diversity of the training data set. Through data enhancement, it is possible to better cope with various complex scenarios and improve the recognition ability of the deep learning model for insulator strings.

7. Use of the method according to any one of claims 1 to 6, characterized in that Use drones equipped with full-frame cameras to collect oblique photogrammetry data, build a three-dimensional model of the transmission line and its surrounding objects, and combine it with deep learning algorithms to achieve automatic identification of insulator strings.

Images