Visual detection algorithm of loose fasteners under the subway
Enhancing the adaptability and accuracy of detection algorithms in subway undercarriage inspection robots is of great significance for improving the intelligent operation and maintenance capabilities of subway.In response to the issues such as stringent shooting conditions required by existing fastener detection algorithms and high demands for anti-loosening line marking,a fastener loosening detection algorithm applied to subway undercarriage inspection robots was proposed.Firstly,when the fasteners in the image were complete and the anti-loosening lines were not significantly obscured,an improved YOLOv5 object detection network was utilized to identify each fastener target.And the fastener targets were divided into three categories.Subsequently,the DeepLabv3_plus image segmentation network was employed to extract the contours of the anti-loosening lines.And these lines were converted into binary images.The angle difference between two anti-loosening lines,the extreme values of the internal angles of the triangle formed by the centroids of three anti-loosening lines,and the area ratio of multiple anti-loosening lines to their minimum bounding rectangle for the three types of fasteners (bolts,nut screws,and metal pipe joints) were calculated after noise reduction.By comparing the calculated values with predetermined thresholds,loosening judgments were made.Finally,the actual loosening conditions within the detected images were tallied,and binary confusion matrices for each type of fastener were formulated.Evaluation metrics were calculated and analyzed,compared with other algorithms,and the algorithm was validated through application to actual subway undercarriage fasteners.The results show that the MAP@0.5 value of the improved YOLOv5 object detection network model increases from 0.877 to 0.911.The MIoU value of the DeepLabv3_plus image segmentation network model reaches 0.950.The MCA values for the 658 fasteners detected by the loosening judgment are 0.907,0.959,and 0.888,which can avoid loose fasteners missing detection.The accuracy rate of various fasteners on the actual subway undercarriage by the algorithm is more than 90％.The feasibility of the improved target detection algorithm,the applicability of the image segmentation network and the reliability of the loosening judgment algorithm are proved by experiments.This algorithm can provide significant technical support for the intelligent operation and maintenance of subway undercarriage fastener loosening detection.
万方数据
维普
