Two-Stage Railway Worker’s PPE Detection Based on Faster-YOLOv9 and ShuffleNetV2

Ensuring railway workers utilize personal protective equipment (PPE) is critical for safety in heavy-haul railways, where complex tasks such as working in busy sections pose significant hazards. Existing detection methods are challenged by high computational complexity and large parameter sizes, hindering efficient monitoring. This paper proposes a lightweight two-stage detection framework. In the first stage, Faster-YOLOv9, which integrates FasterNet with YOLOv9 and incorporates High-level Screening-Feature Fusion Pyramid Networks (HS-FPN), detects workers within railway scenes and crops the corresponding image patches. These patches are then fed into a ShuffleNetV2-based classification model in the second stage to determine PPE compliance. Faster-YOLOv9 employs partial convolution to reduce parameters and computational complexity while maintaining high accuracy. Experimentally, compared to the original YOLOv9, it achieves a 29.8% reduction in parameters, 22.2% lower computational complexity, a 29.9% smaller model size, and 29.5% faster training speed. The ShuffleNetV2 model attains 96.5% accuracy in PPE classification with high computational efficiency. This framework enables efficient and precise detection of PPE compliance for railway workers in challenging heavy-haul environments.