Railway Fastener Detection Based on BAM and DWR Attention Mechanisms

Manual inspection of rail fasteners remains inefficient and insufficient to cover all operational environments, particularly given the complexity and vast extent of railway networks. To address these limitations, an intelligent defect detection framework based on UAVs is developed, with a focus on challenges such as small object scale, complex backgrounds, and data imbalance. The proposed architecture incorporates the BAM and an enhanced C3k2-DWR module within the YOLO11 framework. BAM improves attention to salient regions, while C3k2-DWR enhances multi-scale contextual representation through dilated convolutions and residual connections. Evaluation on high-speed railway image datasets yields a mAP50–95 of 0.913, with inference latency maintained below 40 ms, demonstrating high detection accuracy, robustness, and suitability for real-time deployment in UAV-based railway inspection systems.