A fusion method based on 1D vibration signals and 2D images for detection of railway surface defects

Abstract

Rail surface defects detection is a significant and challenging task to ensure the safety and efficiency of railway transportation. Due to the variety of shapes and sizes of rail surface defects, the existing methods based on a single-source data have low detection accuracy, high false alarm and high miss rate. In this paper, a fusion detection method is proposed based on one-dimensional (lD) vibration signals and two-dimensional (2D) images. Firstly, an Improved Sparse Auto Encoder (ISAE) is used to extract the features of vibration signals. Then Linear Discriminant Analysis-Support Vector Classifier (LDA-SVC) is utilized to diagnose rail surface defects by feature dimension reduction. At the same time, combined with YOLOv5, defects are located and detected based on 2D images. Finally, the results of the two models are fused at the decision-level. The experimental results show that the integration of multi-source information effectively improves the detection accuracy and provides maintenance guidance for the railway department.