Power transmission line power line identification method based on unmanned aerial vehicle laser point cloud

Abstract

The invention provides a power transmission line power line identification method based on unmanned aerial vehicle laser point cloud, and the method employs an unmanned aerial vehicle laser radar to collect the three-dimensional point cloud data of a power transmission line, and achieves the automatic identification of a power line in combination with a deep learning technology. The method comprises the specific steps of data acquisition, preprocessing, three-dimensional modeling and power line automatic identification. A multi-rotor unmanned aerial vehicle carries a laser radar to carry out strip-shaped surrounding flight and collect data of a power transmission line and the periphery of the power transmission line, and after data preprocessing and three-dimensional reconstruction, a PointNet + + deep learning network is used to carry out automatic identification on a power line of the power transmission line, so that the identification efficiency and accuracy can be improved, and compared with traditional manual marking, time consumption and errors are greatly reduced.

Description

Technical Field

[0001] The present invention belongs to the field of laser point cloud data three-dimensional modeling and component extraction, relates to a three-dimensional model of a power transmission line based on point cloud data, and uses PointNet++ deep learning technology to identify power lines of the power transmission line. Background Art

[0002] Transmission lines are important national infrastructure that delivers electricity for daily life and industrial production. It is crucial to ensure the normal operation of high-voltage and low-voltage transmission lines to ensure reliable power supply. Transmission lines are composed of poles, power lines and other supporting facilities. Among them, power lines, as the main body of transmission lines, bear the main function of transmitting electricity. However, due to the long span of transmission lines, long-term exposure to the surrounding environment and the influence of environmental factors such as storms, rainfall and floods, regular monitoring of important facilities such as power lines is of great significance to maintaining the safe and efficient operation of the power grid.

[0003] Traditional power transmission line inspections are all done manually, which has disadvantages such as long time consumption, low efficiency, and low accuracy. Currently, non-contact, large-scale data acquisition methods represented by remote sensing technology are gradually being applied to power inspection work. Light Detection and Ranging (LiDAR), also known as laser scanning technology, provides dense 3D point cloud data and can reconstruct complex 3D details of objects in the scene. At the same time, considering the long length of transmission lines and the complex terrain in the area, drones have become the most convenient platform for data collection required for inspection work. In recent years, with the continuous improvement of the accuracy of laser scanning technology, drone-mounted laser radar scanning equipment has been widely used in the power industry and has become an important technical support for regular inspection, maintenance and discovery of safety hazards of transmission lines.

[0004] After acquiring massive point cloud data of transmission lines, a highly restored three-dimensional model can be further reconstructed, making it easier to locate and identify damaged power facility components. Compared with aerial images, the geometric structure of three-dimensional objects can be further extracted from the point cloud data obtained by laser scanning to achieve detailed three-dimensional reconstruction of power lines. There are many common methods for modeling based on point cloud data, such as interpolation methods, which interpolate point cloud data to generate continuous surface models. Common algorithms include Moving Least Squares (MLS) and spline interpolation. Another example is volume modeling methods, which constructs a three-dimensional voxel model through point cloud data. Common algorithms include voxelization and octree.

[0005] On the basis of constructing a three-dimensional model of the transmission line in the early stage, there are certain challenges in how to further extract the power lines from the dense point cloud obtained by the lidar. At present, there are three main different types of power line extraction methods, one is based on analysis and evaluation, the second is based on supervised classification, and the third is based on deep learning. However, due to reasons such as instruments, environmental conditions, and objects in the scene, various noises and point clouds of unnecessary objects are inevitably present in the data; at the same time, the structure of the power line itself is relatively slender, and the acquired point cloud data itself has certain deficiencies. In addition, there are various types of power lines in the transmission line, including single conductors, lightning conductors, and split conductors. This series of problems will make the power line extraction process more difficult. In order to solve this series of problems, it is objectively necessary to develop a new power line identification method for transmission lines based on drone laser point clouds, which can achieve robust extraction of power lines. Summary of the invention

[0006] The technical solution provided by the present invention is a power line identification method for transmission lines based on drone laser point cloud, which aims to collect three-dimensional point cloud data around the transmission lines through drone-mounted laser radar, and combine deep learning technology to realize automatic identification of power lines, thereby avoiding the tedious process of traditional manual labeling and improving identification efficiency and accuracy.

[0007] A multi-rotor drone equipped with a laser radar device is used for data collection. The drone follows the planned ribbon flight path and circles the power transmission line to obtain data from multiple angles and heights to ensure full coverage of the point cloud data. In order to improve the accuracy of the collected data, control points are laid out on the ground and accurately measured by GNSS equipment to provide reference standards. After completing the data collection, the point cloud data is first converted into common formats, such as LAS, LAZ, etc., for subsequent processing. Then, the point cloud data is preprocessed, including filtering the noise points in the point cloud using statistical filtering methods and removing outliers using outlier detection algorithms. In order to improve the compatibility of point cloud data with other geographic data sets, its coordinates need to be converted into a unified geographic coordinate system, and the preprocessed point cloud data is 3D modeled. Color information is given to the point cloud data by fusing optical images to enhance the intuitiveness and visualization of the model. For the missing parts in the point cloud data, especially the power line part, curve fitting and interpolation methods are used to complete them, and triangular mesh reconstruction or Poisson surface reconstruction is used to generate a complete 3D model. Finally, the PointNet++ network in deep learning is used to automatically identify the power lines. First, some point cloud data are manually annotated to train the PointNet++ model, and data enhancement is performed to improve the robustness of the model. The identified power line points are further processed by clustering and filtering methods to ensure the continuity and accuracy of the recognition results. Finally, the recognition effect is evaluated by overall accuracy (OA) and intersection over union (IoU).

[0008] The technical solution provided by the present invention is: a method for identifying power lines of power transmission lines based on UAV laser point cloud. The specific process is as shown in the attached figure. Figure 1 As shown, it mainly includes the following steps:

[0009] 1) UAV LiDAR point cloud data collection

[0010] 2) Transmission line point cloud data preprocessing

[0011] 3) 3D point cloud modeling of transmission lines

[0012] 4) Automatic identification of power lines BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 Flow chart of the method of the present invention.

[0014] Figure 2 This is the flow chart of UAV radar point cloud data collection.

[0015] Figure 3 This is a diagram of the UAV system composition.

[0016] Figure 4 An overview map of the study area.

[0017] Figure 5 A drone route map.

[0018] Figure 6 This is a map of the distribution of ground control points.

[0019] Figure 7 are the ground control point coordinates.

[0020] Figure 8 Point cloud map of the study area.

[0021] Fig. 9 This is a point cloud image of power lines.

[0022] Fig.10 Complete the graph for power line fitting.

[0023] Fig.11 This is a model diagram of the transmission line and surrounding terrain.

[0024] Fig.12 This is the PointNet++ network structure diagram.

[0025] Fig.13 This is a power line identification diagram. DETAILED DESCRIPTION

[0026] (1) UAV LiDAR point cloud data collection

[0027] The specific process of UAV LiDAR point cloud data collection is as follows: Figure 2 , select a multi-rotor drone equipped with a laser radar. This method takes the DJI M350 RTK equipped with a DJI Zenmuse L2 as an example (see Appendix Figure 3 ), the study area is taken as an example in Yushan Town, Kunshan City, Suzhou City, Jiangsu Province (approximately 120.89° east longitude, 31.35° north latitude) (Appendix Figure 4 ). Use ribbon-shaped circumferential flight to collect point cloud data of the transmission line and its surrounding features (see Appendix Figure 5), the strip surround can collect data from different angles and heights, which helps to obtain all-round and multi-angle information. First, investigate the topography, meteorological conditions, obstacles and flight restricted areas in the mission area to ensure flight safety. Then use the planning software to design the route, generate accurate flight paths and altitude information, avoid complex terrain, obstacles and no-fly zones, and consider the optimal path for flight distance and energy consumption. It is also necessary to plan the flight route according to the surrounding environment, and set flight parameters such as lens photography parameters, flight altitude, and image overlap. Then set specific related parameters, such as lens photography parameters, altitude, overlap and other information. At the same time, aerial photography target boards are laid out on the ground as ground control points (attached Figure 6 ), the control point should have a wide field of view and avoid the influence of multipath effect. And use GNSS equipment to accurately measure the plane coordinates and elevation information of the target (see Appendix Figure 7 ), providing a reference standard for the collected point cloud data to improve the accuracy of the model.

[0028] (2) Transmission line point cloud data preprocessing

[0029] After the data collection is completed, the collected raw point cloud data is converted into common point cloud formats (such as LAS, LAZ, PLY, etc.) for subsequent processing. In order to improve the quality of the point cloud data, it needs to be preprocessed. First, the noise points in the point cloud are filtered using a statistical filtering method to reduce random interference in the environment. Secondly, outlier detection algorithms (such as the DBSCAN clustering algorithm) are used to remove outliers, thereby improving the overall consistency and accuracy of the data. Finally, the coordinate system of the point cloud data is converted from the device coordinate system to a unified geographic coordinate system for integration and compatibility with other data sets. The point cloud visualization results are shown in the attached figure. Figure 8 .

[0030] (3) 3D point cloud modeling of power transmission lines

[0031] The pre-processed point cloud data is used for 3D modeling. First, the point cloud data is given color information by fusing optical images, thereby improving the visualization of the model and making the model more intuitive. In the modeling process, the missing parts in the data, such as the power line part (attached Fig. 9 ), curve fitting and interpolation methods are used to complete the model to ensure the integrity of the model. Specifically, the least squares method is used to fit the power line curve to generate a smooth power line curve (see Appendix Fig.10 ). The mathematical expression of the parabola-based power line fitting is:

[0032] z=ax 2 +by 2 +cxy+dx+ey+f Formula (1)

[0033] Among them, a, b, c, d, e, and f are the coefficients that need to be fitted, which are obtained using the least squares method. The objective function is:

[0034]

[0035] Where N represents the number of data points, (x i ,y i ,z i ) represents the coordinates of the i-th data point.

[0036] Subsequently, the discrete point cloud data is connected into a mesh or polygonal surface using triangular mesh reconstruction technology or Poisson surface reconstruction technology to generate a complete 3D model. After the modeling is completed, the model is further processed using Laplace smoothing and bilateral filtering methods to eliminate sharp edges and noise and improve the overall quality of the model. The results of the 3D modeling of the transmission line point cloud are shown in the attached figure. Fig.11 .

[0037] (4) Automatic identification of power lines

[0038] In the automatic identification stage of power lines, the PointNet++ network is used to process and identify point cloud data. The PointNet++ network structure is shown in the attached figure. Fig.12 First, some of the collected point cloud data are manually annotated as training data sets to train the PointNet++ model. In order to enhance the robustness of the model, the training data is enhanced, including posture enhancement and interpolation enhancement methods, to increase the diversity of the data. After the training is completed, the trained model is used to automatically recognize new point cloud data. The recognition results are shown in the attached figure. Fig.13 . During the recognition process, the identified power line points are processed by clustering and filtering methods to eliminate misidentified point clouds and ensure the continuity and accuracy of the recognition results. Finally, the recognition effect is evaluated by indicators such as Overall Accuracy (OA) and Intersection over Union (IoU) to ensure the reliability of the model. Overall accuracy refers to the proportion of samples correctly classified by the model on the entire data set, which is used to measure the overall classification performance of the model. The intersection over union refers to the ratio of the intersection and union of the segmented area predicted by the model and the true label area. The mathematical expressions of OA and IoU are:

[0039]

[0040] Among them, TP (True Positive) refers to the number of positive samples correctly identified; FP (False Positive) refers to the number of falsely reported negative samples; TN (True Negative) refers to the number of missed positive samples; FN (True Negative) refers to the number of negative samples correctly identified.

[0041] In this method, the overall accuracy and intersection-over-combination ratio of power line identification are 89.61% and 87.75%, respectively.

Claims

1. A method for identifying power lines of power transmission lines based on UAV laser point cloud, characterized in that: Collect the UAV LiDAR point cloud data of the power transmission line and its surrounding objects, build a 3D model of the power transmission line and its surrounding objects after preprocessing, and then automatically identify the power line with deep learning, including the following steps: (1) UAV LiDAR point cloud data collection (2) Transmission line point cloud data preprocessing (3) 3D point cloud modeling of transmission lines (4) Automatic identification of power lines.

2. The method according to claim 1, characterized in that: The UAV laser radar point cloud data collection includes: using a multi-rotor UAV equipped with a laser radar to collect data on the power transmission line and its surrounding objects by flying in a belt-shaped circle, thereby obtaining comprehensive point cloud data. To ensure the high accuracy of the data, an aerial target plate is laid on the ground as a ground control point, and the GPS coordinates and elevation information of these control points are measured by a GNSS receiver to facilitate the subsequent verification of the accuracy of the UAV data.

3. The method according to claim 1, characterized in that: The transmission line point cloud data preprocessing includes: firstly, converting the format, then using statistical filtering method and outlier detection algorithm to remove noise points and outliers. Then, the point cloud data needs to be converted into coordinates so as to be compared with the ground control points to verify the accuracy and integrity of the data.

4. The method according to claim 1, characterized in that: The three-dimensional modeling of the point cloud of the power transmission line includes: coloring the point cloud and adding its color attribute information to make the three-dimensional model more intuitive. Subsequently, the power line fitting algorithm based on the three-dimensional parabola is applied to complete and fit the point cloud data to ensure the integrity of the power line data. Through these steps, a complete three-dimensional model of the power transmission line and its surrounding objects is constructed, and the model is finely reconstructed to ensure the accurate expression of the features of the objects.

5. The method according to claim 1, characterized in that: The automatic identification of power lines includes: using a deep learning method to automatically identify power lines. This process uses deep learning image recognition technology to achieve accurate identification of power lines through training with a large amount of data. The recognition algorithm is designed to automatically identify power lines, and the overall accuracy (OA) and intersection over union (IoU) are used to evaluate the accuracy of the recognition effect.

6. The method according to claim 5, characterized in that The recognition process includes: introducing data enhancement operations in the recognition process, including posture enhancement and interpolation enhancement methods. These data enhancement operations effectively increase the diversity of the training data set, thereby improving the generalization ability of the deep learning model and further improving the accuracy of power line recognition.

7. Use of the method according to any one of claims 1 to 6, characterized in that Drones are used to collect lidar point cloud data to build a three-dimensional model of the transmission line and surrounding objects, and deep learning methods are used to realize automatic identification of power lines.

Images