Foreign Object Detection Algorithm Based on Multi-scale Convolutional Network

Abstract

In order to detect foreign matter invading the track and prevent the intrusion of foreign matter from causing railway safety accidents, the detection algorithm of foreign matter intrusion on the track is studied. Aiming at the problem that the rail transit scene is complex and the obstacle scale changes in the acquired image information, this research proposes a multi-scale target detection algorithm based on the YOLO (you only look once) algorithm. First, an adaptive feature fusion module is designed to make the feature maps used for detection have strong semantic information at various scales; then, a new loss function is designed to alleviate the problem of uneven sample distribution and optimize the training process. Experiments show that the algorithm has obvious advantages in multi-scale detection, which not only improves the accuracy of target detection, especially the accuracy of small targets, but also does not significantly increase the inference time and the amount of parameters, and has high real-time performance.