RM-YOLOv8-n: A Lightweight and High-precision Network for Rail Surface Defect Detection

The surface condition of rail is important for ensuring the safe and stable running of railway vehicles, so real-time defect detection of rail surfaces is essential. However, manual inspection and mainstream non-destructive surface detection methods make meeting the accuracy and efficiency requirements difficult. This paper proposes a rail surface defect detection method: reversible multi-scale you only look once-v8 (RM-YOLOv8-n) on top of improved YOLOv8-n, which can identify rail surface defects more accurately and quickly with fewer parameters. First, the backbone is reconstructed by using the idea of reversible column networks (RevCol) to complete feature extraction more efficiently. Second, the multi-section block with attention and pooling (MSAP) module is designed to improve attention to defects and reduce noise interference in the process of feature fusion. In addition, ghost convolution with shuffle (GSConv) is introduced to reduce the computational complexity in the process of down-sampling and further optimize the information interaction. Experiments on the rail surface defect dataset show that our model achieves the highest mAP@O.5 of 77.4% while reducing computational consumption by 37%