摘要
Abstract
In view of such problems as time and labor consuming as well as safe hazard in the patrol inspection of the insulator defect by artificial portable instrument of electrified railway and for identifying insulator defect accurately and quickly by using intelligent image processing technology,a kind of insulator defect identification algorithm based on lightweight object detection is proposed.The YOLO(you only look once)v3 model is taken as the base,the lightweight network ShuffleNetv2(shuffle netv2)is used as the backbone network,and spatial attention mechanisms(SAM)is added to enhance the feature extraction ability of the model.K-means ++ is used as anchor frame cluster-ing algorithm,and distance-intersection over union loss non-maximum suppression(DIoU-NMS)is used as boundary loss function to improve the performance of the algorithm further.The experimental results show that the Recall,Precision,F1-Score(F1),average precision(AP)and frames per second(FPS)of the proposed algorithm reach 91.52%,92.10%,91.81%,92.02%and 52.62 frames/s,respectively,on the common data set CPLID.The above in-dexes on the self-made data set reach 93.12%,92.79%,92.95%,92.47%and 64.38 frames/s,respectively,and the de-tection accuracy and speed reach effective balance.关键词
绝缘子/缺陷检测/YOLOv3/轻量化/边界损失函数Key words
insulator/defect detection/YOLOv3/lightweight/boundary loss function
