STOE-YOLOv5s: A Hard Hat Wearing Detection Algorithm Based on Improved YOLOv5s

To address the existing issues of missed detection and false detection of hard hats by mainstream detection models in complex environments, an improved hard hat detection algorithm, STOE-YOLOv5s based on YOLOv5s, has been proposed. Firstly, enhance the detection header structure, add multi-scale target detection layers, and improve the detection capability for small targets; Secondly, a BottleNeck Transformer (BoT3) module based on self-attention is added to the backbone network to enhance the model's ability to extract global features; Then, to solve the problem of positive and negative sample matching, a Simplified Optimal Transport Assignment (SimOT A) strategy is used to obtain the optimal matching result in the global range; Finally, an Efficient Multi-Scale Attention (EMA) mechanism is fused into the C3 module of the network neck to enhance the model's ability to capture features-rich areas. Experimental results on the hard hat dataset show that the mAP of the improved algorithm reaches 96.0%, and the recall rate reaches 91.8%. Compared with the original YOLOv5s algorithm, it has increased by 1.7 and 2.9 percentage points respectively, proving the effectiveness and advancement of the improved algorithm.