Missing Small Bolt Detection on High-speed Train Using Improved Yolov5

Tiny bolts are widely used on high-speed trains, playing an important role in fixing train components. However, because of the complex running environment of trains, missing bolts occur from time to time and may cause traffic accidents, resulting in property damage and, in serious cases, endangering the lives of the occupants. Therefore, it is essential to detect missing bolts on high-speed trains. The bolts discussed in this paper are generally located on the underside of high-speed trains and their small size makes detection more difficult. In this paper, we first expand the dataset, and then add the Attention module and Transformer based on YOLOv5, and change the FPN of YOLOv5 to BiFPN, fuse the features of different layers using different weights, and crop the high-resolution original image during training and testing, and finally return to the original image. Our method eventually achieves 95.3% AP, effectively improving the detection accuracy.

I. Introduction

With the development of rail transport, more and more high-speed trains are being put into service. On high-speed trains, bolts are much widely applied to transfer, connect and fasten many kinds of parts. but because of the complex running environment, bolt defects occur from time to time, threatening the safety and stability of high-speed trains. Figure 1 shows the normal bolts and missing bolts on high-speed train. In addition, the scattered distribution and tiny size of these parts makes detection of defects difficult.¹