An improved YOLOv5s for protective gear detection
Xinghao Cheng, Tao Lü · 2022 7th International Conference on Intelligent Computing and Signal Processing (ICSP) · 2022
Live working refers to a method of maintenance and test on high voltage electrical equipment without power failure, which requires direct contact with wires above 10kV. At present, relying on manual inspection of protective gear is time-consuming and laborious, and there are still many deaths caused by negligence at work. In this paper, we applies YOLOv5s to this task. However, due to the similar color of the detected targets and the small targets, many false detections and missed detections occurred in the results. Aiming at these problems, we propose an improved YOLOv5s-CSBN(CAM, SAM, BiFPN, New detection scale). First, We propose a new method of embedding SAM(spatial attention module) and CAM(channel attention module) to the network, with the best detection performance improvement; Second, we add a new detection scale and introduce BiFPN instead of PANet. Finally, experimental data show that the [email protected]% of YOLOv5S-CSBN is 2.2% higher than that of Yolov5s in our data set, with significantly reducing the false detection and missed detection caused by similar colors and small targets.