Research on small foreign body detection of contact net based on improved YOLOv10

Abstract

High-speed railway contact line is the core facility of traction power supply for electrified railway, and its safe operation is directly related to the reliability of train operation. Aiming at the problems of low detection accuracy and poor real-time performance of small foreign bodies in contact net scene caused by small scale, weak characteristics, and easy interference from complex background, a high-precision detection method of small foreign bodies in contact net based on improved YOLOv10 was proposed. Firstly, a special contact net data set containing typical small foreign objects such as plastic bags, kite lines, and bird nests was constructed. Secondly, a lightweight Coordinate Attention (CA) module was embedded in the YOLOv10 backbone network to strengthen the perception of small target features, and the Bidirectional Feature Pyramid Network (BiFPN) was used to replace the original PANet to optimize multi-scale feature fusion. Then, a Small Object Focal Loss is designed to alleviate the problem of sample imbalance. Finally, comparative experiments were carried out on self-made datasets. The results show that the proposed model has significant improvement compared with the existing methods.