YOLOv5-NCM for noise barrier defect detection using UAV-based images in high-speed railway

Noise barriers are essential for mitigating railway noise and preventing foreign object intrusion onto tracks. During operation, they are vulnerable to environmental factors that can cause structural defects, including surface damage, rusted column and deteriorated mortar layer, which compromise noise insulation and present safety hazards. Traditional manual inspection methods are inefficient, costly, and subjective, making them unsuitable for large-scale noise barrier assessments. This paper proposes a detection method that utilizes unmanned aerial vehicle (UAV) imagery through an enhanced YOLOv5 network, named YOLOv5-NCM, to improve defect detection accuracy and efficiency in noise barriers. First, to address the lack of noise barrier defect datasets, the Noise Barrier Defect Detection (NBDD) dataset was constructed using UAV imagery of noise barriers along high-speed rail lines. Second, an NC3 module was designed to replace the C3 module in YOLOv5’s backbone network. Additionally, a Multiple Attention Module (MAM) was introduced to enhance the model’s feature-learning capabilities. Experimental results show that YOLOv5-NCM improves detection accuracy by 0.4% on the NBDD dataset compared to the original YOLOv5. YOLOv5-NCM demonstrates superior performance in identifying noise barrier defects, offering an efficient solution for noise barriers defect detection in high-speed railway.