Improved YOLOv5 Algorithm for Helmet Wearing Detection
Benqi Liao, Ting Chen, Pan Xue · 2023
In order to solve the problem that manual detection of helmet wearing in field operation is not only time-consuming and laborious, but also unable to achieve real-time rapid monitoring, a helmet wearing detection algorithm with CBAM lightweight is proposed. The algorithm is based on the YOLOv5 algorithm, firstly, the backbone network of YOLOv5 is replaced by a lightweight network MobileNetV2 to reduce the number of parameters of the algorithm to meet the demand for rapid detection; secondly, channel and spatial attention mechanisms are added to the algorithm, which enables it to extract features from the input image information sequentially and fully in channel and space. The experimental results show that the improved algorithm achieves a mAP of 94.8%, a detection speed of 42FPS, and a model size of only 1.7MB, which is 5.7% higher in mAP, 13FPS faster in detection speed, and about 8 times smaller in model size compared with YOLOv5, thus realizing more accurate, faster, and real-time helmet wearing detection. The method is applied in actual production operation, which compensates for the time-consuming and laborious shortcomings of manual supervision at the operation site.