Smaller Target Detection Algorithms Based on YOLOv5 in Safety Helmet Wearing Detection

Abstract

In many construction scenarios, the frequency of workers' safety accidents has gradually increased. The main reason is that workers do not wear safety helmets. Safety helmets are a kind of safety guarantee for workers, which can effectively prevent workers' head injuries. Therefore, wearing a helmet correctly can effectively reduce the frequency of safety accidents. With the continuous development of the field of computer vision, deep learning has made remarkable achievements in safety helmet wearing detection. For safety helmet wearing detection, we can detect whether workers in construction sites wearing safety helmet by establishing a detection model. Aiming at the shortcomings of the existing helmet detection algorithms, it is difficult to detect small targets, occluded targets, and dense targets. This paper proposes a helmet wearing detection algorithm based on YOLOv5. Add a smaller target detection layer P2 to improve the efficiency of small target detection, and we use the FReLU activation function to replace the original SiLU activation function. Finally, the coordinate attention is added to the backbone network to enhance the network's learning of the details of the safety helmet target. The experimental results show that under the SHWD dataset, compared with the original model, the average accuracy of the model reaches 95.2%, an improvement of about 1.9%. The improved algorithm has good accuracy and practicability for the helmet detection task, and can better meet the actual needs of the construction site.