An Improved YOLOv8n with Multi Scale Feature Fusion for Real Time and High Precision Railway Track Defect Detection

Zhihong Zhang^1,2*Liling Zhang¹Xin Lu¹Tingting Ma¹Feng Huang¹Sheng Zhong²

• ¹Guangzhou Institute of Metrology and Testing Technology, Guangzhou, China
• ²Huazhong University of Science and Technology, Wuhan, China

Railway transportation has become increasingly vital for modern urban and intercity mobility, yet the growing scale and operational intensity of rail networks have made defect detection in railway tracks a critical concern. Traditional inspection methods such as manual, ultrasonic, eddy current, and magnetic flux leakage testing suffer from limitations in accuracy, efficiency, or adaptability to complex environmental conditions. To address these challenges, this study proposes an enhanced defect detection framework based on an improved YOLOv8 algorithm, specifically designed for small targets and complex backgrounds. The improvements include: (1) an AVCStem module with variable convolution kernels to dynamically adapt to defects of different shapes and scales; (2) an ADSPPF module using multi-scale pooling and multi-branch attention mechanisms to retain fine features across scales; and (3) a MSF module for enhanced multi-scale feature fusion with partial convolution and hierarchical feature alignment. Experimental results on a real-world track defect dataset demonstrate that the proposed model achieves a significant increase in detection precision (90.2%), mAP@0.5 (90.2%), and mAP@0.5:0.95 (73.2%), while reducing model size to 5.2MB and parameters to 2.45M. Comparative and ablation studies further validate the complementary benefits of each module and the superior performance over existing lightweight detectors. The proposed model offers a robust, accurate, and efficient solution for real-time railway defect detection, with strong potential for deployment in edge AI devices and mobile inspection robots.