Railway fastener defect detection based on improved YOLOv5 algorithm
Zhitong Su, Kai Han, Wei Song, Keqing Ning
Abstract
The health condition of fasteners is the key to ensure the normal operation of track vehicles. At present, manual inspection is used for track maintenance. Aiming at the inaccuracy and inefficiency, a method of track fastener defect detection based on improved YOLOv5 is proposed. Firstly, the size of the target box of fastener defect is analyzed by K-means algorithm to determine the size of the top priority box. Secondly, combining attention mechanism with multi-scale fusion, this paper analyzes the small objects of railway fasteners. The method is applied to the railway fastener defect data set. Experimental results show that: The average accuracy of the improved YOLOv5 model improved by about 0.9% to 96.1%, allowing for accurate and rapid identification of fastener defects.