Image Recognition of Bird Pecking Defects in Composite Insulators Based on Improved HED Network

Abstract

AbstractBird pecking damage may affect the performance of the composite insulator in service by increasing the probability of intense surface partial discharge activity, which directly threatened the safe operation of the power grid. Using reasonable composite insulator damage identification technology to identify damaged insulators in time is vital to preventing transmission line accidents. To recognize composite insulator damage, a bird pecking defect image recognition method based on an improved HED network is proposed. Firstly, as a result of preprocessing the collected images, the median filter algorithm smooths out the images and the Retinex algorithm reduces the influence of uneven illumination and improves the accuracy of composite insulator damage image recognition. Secondly, based on the semantic segmentation method, a mask containing semantic information of composite insulator target area is obtained, and the sample is divided into background area and composite insulator target area. Finally, the edge features of bird pecking defect of composite insulator are extracted by HED image recognition method, and the bird pecking damage recognition of composite insulator is completed. The experimental results demonstrate that the proposed method is effective in identifying bird pecking damage of composite insulators, as well as improving the efficiency of power grid operations.KeywordsComposite insulatorBird pecking damageImage recognitionSemantic segmentationHolistically-nested edge detection network