Helmet Detection Recognition Algorithm Based on Improved Yolov5

The existing helmet detection algorithm is mainly based on a single-stage object detection algorithm, which has high detection speed and can achieve the requirement of real-time detection. Still, the accuracy of detecting small objects and objects with obstacles is no high. Based on this, this paper proposes a helmet detection recognition algorithm based on improved YOLOv5. The SE attention mechanism has the advantages of low complexity, few parameters and low computational effort. Firstly, the SE attention module is added to the Yolov5 backbone network to weigh the features of different channels and improve the model's attention to important features. Secondly, using softnms to replace nms in the original network can retain more small target prediction frames, improve the detection accuracy, and improve the model's ability to detect small targets. The experiments show that the average accuracy, precision, and recall of the proposed method reach 96.51%, 95.20%, and 90.08%, respectively, and the real-time performance meets the need for helmet detection in practical work.