CN-120807393-B - Binocular vision-based wire insulator defect detection method and system

CN120807393BCN 120807393 BCN120807393 BCN 120807393BCN-120807393-B

Abstract

The application discloses a binocular vision-based wire insulator defect detection method and a binocular vision-based wire insulator defect detection system, wherein firstly, a multi-view image of an insulator is obtained, and preliminary three-dimensional point cloud data of the insulator are obtained through calculation according to the multi-view image; and then, calling a three-dimensional defect detection model integrated with a plurality of modules such as a multi-scale feature extraction module, a mixed attention module, a feature alignment module, a feature enhancement module and the like, and detecting according to the three-dimensional point cloud data and a two-dimensional image of the insulator to obtain a defect detection result of the insulator. According to the embodiment of the application, the problem of shooting angle is solved by acquiring the three-dimensional image through the binocular vision camera, the dynamic illumination compensation is utilized to cope with illumination change, and the three-dimensional defect detection model is matched to overcome the bottleneck of micro defect detection, so that the accuracy of defect detection is effectively improved, and the high-precision insulator defect detection is realized.

Inventors

LIN JIANPEI
GUAN YAOXIAN
YIN QIFEI
ZOU HAIBO
TAN JIEXIA
LI YUNLIN
QU CAIJUAN

Assignees

江门明浩实业集团有限公司

Dates

Publication Date: 20260505
Application Date: 20250604

Claims (9)

1. The utility model provides a wire insulator defect detection method based on binocular vision which is characterized in that the method comprises the following steps: Acquiring a multi-view image of an insulator, and calculating to obtain preliminary three-dimensional point cloud data of the insulator according to the multi-view image; performing point cloud denoising and dynamic illumination compensation processing on the preliminary three-dimensional point cloud data to obtain optimized three-dimensional point cloud data; invoking a pre-trained three-dimensional defect detection model, and detecting according to the three-dimensional point cloud data and the two-dimensional image of the insulator to obtain a defect detection result of the insulator; The three-dimensional defect detection model comprises a multi-scale feature extraction module, a mixed attention module, a feature alignment module, a feature enhancement module, a defect candidate generation module and a defect result generation module, wherein the multi-scale feature extraction module performs feature extraction and transformation processing of different scales on the three-dimensional point cloud data to obtain multi-scale geometric features, the mixed attention module performs attention feature extraction on the multi-scale geometric features in combination with the two-dimensional images to obtain attention weighted features, the feature alignment module performs feature alignment processing of the attention weighted features under different view angles to obtain aligned features, the feature enhancement module performs coding, feature transformation and feature mapping processing on the aligned features to obtain enhanced features, the defect candidate generation module performs feature extraction according to the enhanced features to obtain candidate defect areas, and the defect result generation module performs defect positioning according to the candidate defect areas to obtain defect detection results; The multi-scale feature extraction module comprises a multi-scale neighborhood grouping unit, a pooling layer and a multi-layer perceptron, wherein the multi-scale neighborhood grouping unit comprises a plurality of neighborhood radius layers with different scales, each neighborhood radius layer is connected with the pooling layer, and the pooling layer is connected with the multi-layer perceptron.
2. The binocular vision-based wire insulator defect detection method of claim 1, wherein the hybrid attention module comprises a channel attention unit and a spatial attention unit, the channel attention unit comprising a connected characteristic channel layer and a crush stimulus module, the spatial attention unit comprising a connected spatial convolution layer and a pooling layer.
3. The binocular vision-based wire insulator defect detection method of claim 1, wherein the feature enhancement module comprises a neural implicit function unit and a feature fusion unit which are connected with each other, and the neural implicit function unit comprises a coding layer and a plurality of full-connection layers and sinusoidal activation functions which are connected with each other.
4. The binocular vision-based wire insulator defect detection method of claim 1, wherein the calculating the preliminary three-dimensional point cloud data of the insulator according to the multi-view image comprises: calculating pixel intensity difference cost of the multi-view image to obtain a parallax value; Calculating to obtain path accumulated costs in multiple directions according to the parallax value through a multi-direction path cost aggregation strategy; Determining a target parallax value according to the path accumulated cost of the multiple directions; And calculating to obtain preliminary three-dimensional point cloud data of the insulator according to the target parallax value, the camera focal length and the binocular baseline distance.
5. The binocular vision-based wire insulator defect detection method of claim 1, wherein the performing the point cloud denoising and dynamic illumination compensation processing on the preliminary three-dimensional point cloud data to obtain optimized three-dimensional point cloud data comprises: Carrying out statistical filtering treatment on the preliminary three-dimensional point cloud data, and eliminating abnormal points obtained after treatment to obtain denoising point cloud data; Performing self-adaptive histogram equalization processing on the multi-view image according to the denoising point cloud data to obtain an illumination-homogenized multi-view image; and mapping the gray value of the illumination homogenized multi-view image to the denoising point cloud data to obtain optimized three-dimensional point cloud data.
6. The binocular vision-based wire insulator defect detection method of claim 1, wherein the multi-view image of the insulator is obtained according to the steps of: calibrating internal parameters and external parameters of the binocular vision camera, aligning the external trigger signals with the time stamp, and setting a synchronization mechanism of the binocular vision camera to obtain a preset binocular vision camera; And carrying out surrounding type image acquisition on the insulator by the preset binocular vision camera to obtain a multi-view image of the insulator.
7. The binocular vision-based wire insulator defect detection method of claim 6, wherein the defect result generation module comprises an image extraction unit, a detection head and a defect positioning unit which are connected with each other, wherein the defect result generation module is used for performing defect positioning according to the candidate defect area to obtain a defect detection result, and the method comprises the following steps: Extracting a feature image from the candidate defect region by using the image extraction unit; detecting and classifying the characteristic images by using the detection head to obtain defect probability distribution of a plurality of defects; and calculating the confidence coefficient of each defect by using the defect positioning unit according to the defect probability distribution of the defects and the marked defect areas in the three-dimensional point cloud data, and obtaining a defect detection result according to the confidence coefficient.
8. A binocular vision-based wire insulator defect detection system, which is applied to the binocular vision-based wire insulator defect detection method as claimed in any one of claims 1 to 7, comprising: The data acquisition module is used for acquiring multi-view images of the insulator and calculating to obtain preliminary three-dimensional point cloud data of the insulator according to the multi-view images; the data processing module is used for carrying out point cloud denoising and dynamic illumination compensation processing on the preliminary three-dimensional point cloud data to obtain optimized three-dimensional point cloud data; the defect detection module is used for calling a pre-trained three-dimensional defect detection model, and detecting according to the three-dimensional point cloud data and the two-dimensional image of the insulator to obtain a defect detection result of the insulator; The three-dimensional defect detection model comprises a multi-scale feature extraction module, a mixed attention module, a feature alignment module, a feature enhancement module, a defect candidate generation module and a defect result generation module, wherein the multi-scale feature extraction module performs feature extraction and transformation processing of different scales on the three-dimensional point cloud data to obtain multi-scale geometric features, the mixed attention module performs attention feature extraction on the multi-scale geometric features in combination with the two-dimensional images to obtain attention weighted features, the feature alignment module performs feature alignment processing of the attention weighted features under different view angles to obtain aligned features, the feature enhancement module performs coding, feature transformation and feature mapping processing on the aligned features to obtain enhanced features, the defect candidate generation module performs feature extraction according to the enhanced features to obtain candidate defect areas, and the defect result generation module performs defect positioning according to the candidate defect areas to obtain defect detection results; The multi-scale feature extraction module comprises a multi-scale neighborhood grouping unit, a pooling layer and a multi-layer perceptron, wherein the multi-scale neighborhood grouping unit comprises a plurality of neighborhood radius layers with different scales, each neighborhood radius layer is connected with the pooling layer, and the pooling layer is connected with the multi-layer perceptron.
9. The binocular vision-based wire insulator defect detection system of claim 8, wherein the data acquisition module comprises a binocular vision camera.

Description

Binocular vision-based wire insulator defect detection method and system Technical Field The embodiment of the application relates to the technical field of image processing, in particular to a binocular vision-based wire insulator defect detection method and system. Background With the continuous expansion of the power grid scale, the safety and stability of the high-voltage transmission line become the core problems of concern in the power industry. The insulator is used as a key component of the power transmission line, and is mainly used for fixing a wire and providing insulation support between the wire and an electric pole or a tower so as to ensure that current is transmitted in a set path. However, once the insulator is defective, the insulation performance is reduced, serious accidents such as partial discharge, flashover, pollution flashover and the like can be possibly caused, and even in extreme cases, line tripping or large-range power failure can be caused, so that the safe and stable operation of a power grid is seriously threatened. Therefore, the high-efficiency and accurate insulator defect detection method is designed, the state sensing and operation and maintenance efficiency of the power equipment are improved, and the method has important significance for guaranteeing the safety and reliability of a power grid. At present, the inspection of the wire insulator mainly depends on a two-dimensional computer vision technology mode for detection, an unmanned aerial vehicle or a fixed camera is utilized for shooting an insulator image, and then a deep learning model is utilized for defect identification. However, the current two-dimensional image detection method has inherent limitations, the identification effect is influenced by factors such as shooting angles, illumination changes, obstruction interference and the like, the complete space information of the defects is difficult to obtain, meanwhile, the defects of the insulators are generally tiny, and the detection precision and the robustness are insufficient. A new detection method is needed to break through the limitation of two-dimensional visual detection and realize accurate identification and spatial positioning of insulator defects. Disclosure of Invention The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims. The embodiment of the application provides a method and a system for detecting defects of an electric wire insulator based on binocular vision, which effectively improve the accuracy of defect detection and realize high-precision defect detection of the insulator by fusing binocular vision three-dimensional modeling and deep learning technologies of an unmanned plane. According to the method, a multi-view image of an insulator is obtained, preliminary three-dimensional point cloud data of the insulator are obtained through calculation according to the multi-view image, point cloud denoising and dynamic illumination compensation processing are conducted on the preliminary three-dimensional point cloud data, optimized three-dimensional point cloud data are obtained, a pre-trained three-dimensional defect detection model is called, defect detection results of the insulator are obtained through detection according to the three-dimensional point cloud data and a two-dimensional image of the insulator, the three-dimensional defect detection model comprises a multi-scale feature extraction module, a mixed attention module, a feature alignment module, a feature enhancement module, a defect candidate generation module and a defect result generation module, the multi-scale feature extraction module conducts feature extraction and transformation processing on the three-dimensional point cloud data according to the multi-scale feature extraction module to obtain multi-scale geometric features, the mixed attention module conducts attention feature extraction on the multi-scale geometric features in combination with the two-dimensional image to obtain attention weighted features, the feature alignment module conducts feature alignment processing on the non-weighted attention feature, the feature alignment module conducts feature alignment enhancement module to obtain candidate feature enhancement module, and the defect generation module conducts candidate feature enhancement processing on the candidate region enhancement module according to the candidate feature enhancement module, and the defect generation module obtains the candidate region enhancement feature enhancement result. With reference to the first aspect, in an embodiment of the present application, the multi-scale feature extraction module includes a multi-scale neighborhood grouping unit, a pooling layer and a multi-layer perceptron, where the multi-scale neighborhood grouping unit includes a plurality of neighborhood radius layers with different scales, each neighborhood radius layer is con