A Novel Two-stage Method For Detecting The Missing Insulator Caps Based On The Rotated Target Framework

In recent years,with the increasing scale of the smart grid and the rapid increase of transmission line complexity,China’s power grid inspection and maintenance are facing a huge challenge.Insulator cap missing detection is an important part of power grid inspection and maintenance,and automatic detection methods based on computer vision can effectively improve inspection efficiency.However,the existing methods are either vulnerable to the influence of threshold settings or have a limited effect on the detection of images with high background complexity,and there is no effective solution to the problem of mutual occlusion of insulator caps.With the missing insulator cap detection as the research objective,this paper proposes a two-stage automation method based on the rotated detection framework.The main innovations and contributions are as follows:Firstly,an improved method of rotational target detection for insulator caps is proposed.In the application scenario of this paper,the insulators in the UAV images show a certain rotation angle,and if the horizontal rectangular frame detection is used,some of the detection frames will be filtered out when the non-maximum suppression is subsequently used because the intersection ratio threshold of similar detection frames is too large,which leads to the problem of target missing detection.Therefore,this paper proposes an adaptive generation method with rotation angle detection frames to improve the detection accuracy for insulator caps in aerial images without increasing the complexity.Secondly,based on the improved method of rotating target detection above,a twostage insulator cap missing detection method is proposed.When the insulator cap targets are too small and severely occluded from each other,the first-stage method may still have the problem of missing detection.Based on the comprehensive consideration of insulator cap morphological characteristics and spatial coordinate information,a second-stage method based on spatial statistical information is proposed to further optimize the first-stage insulator cap detection results.If the insulator cap is still not detected at the target location,it is determined that there is a cap missing.In addition,in order to meet the practical application needs,a lightweight and acceleration strategy is applied to the detection model to improve the speed of model inference and reduce the number of model parameters while ensuring detection accuracy.In summary,to address the problems of target occlusion and high background complexity,a two-stage insulator missing detection method is proposed which combines a rotational detection framework and a lightweight and acceleration strategy,i.e.,we first achieve insulator cap recognition and then perform insulator cap missing detection based on the recognition results,which effectively improves the missing detection accuracy and detection speed and reduces the number of model parameters.