Research on Small Target Detection Algorithm of Catenary Based on DA-YOLOv4

Abstract

In recent years, computer vision has been greatly developed in the detection of catenary equipment. With its high efficiency and accuracy, it meets the needs of safety detection of catenary equipment in the safe operation of trains. In the catenary monitoring image, some equipment targets are small, which makes it difficult to identify. To solve this problem, this paper proposes an improved small target detection algorithm -DA-YOLOv4. In this method, Dual Attention Network for Scene Segmentation ( DANet ) is integrated into YOLOv4 model. Position Attention Module ( PAM ) and Channel Attention Module ( CAM ) are applied to enhance the attention of feature extraction network to small targets from two aspects of spatial location and feature channel. The context information is fully utilized to solve the problems of difficult feature extraction and low recognition rate of small targets. Experiments show that the DA-YOLOv4 algorithm can effectively improve the detection effect of small targets in the catenary, and the average detection accuracy on the catenary data set is 77.6 %, which is 4.7 % higher than that of the YOLOv4 network.