Currently, railway maintenance safety accidents occur frequently, and existing railway safety supervision schemes are inefficient and difficult to effectively ensure the safe conduct of maintenance activities. A protection equipment detection scheme for railway maintenance workers based on the improved YOLOv5 model was proposed, which realized real-time detection of safety helmets and reflective vests worn by railway maintenance personnel. To address issues such as missed detections, false positives, and inaccurate detections caused by the complex background of railway scenes, a spatial reconstruction unit (SRU) and an efficient channel attention(ECA) module was introduced. These modules were employed to enhance the network's feature expression ability in both the spatial and channel dimensions, effectively filtering out useful features and removing redundant ones. Additionally, weighted intersection over union (W-IOU) loss function was introduced to reduce the impact of low-quality samples on training, thereby improving the convergence ability of model. Experimental validation result shows that the model achieves the average accuracy of 94.3%, the precision of 94.9%, and the recall of 94.7%, which represent the improvements of 2.2%, 2.5%, and 1.6%, respectively, over the baseline network. The inference speed reaches 34.5 FPS, allowing for real-time detection in actual maintenance environments.