Substation helmet detection based on improved YOLOX-S algorithm

Abstract

The improved YOLOX-S algorithm is proposed for the detection of small helmet targets based on an improved YOLOX-S algorithm for the detection of helmets worn by relevant personnel in hazardous scenarios in substations. First, the ECA attention mechanism is introduced into the CSPLayer structure in YOLOX-S to direct the model to pay more attention to channel features of small target information and enhance the model's ability to utilize useful features. Secondly, the addition of the ConvNext Block module after the three feature layers of the backbone feature extraction network to enhance the model's ability to exploit useful features. Finally, the weighted feature fusion mechanism of BiFPN is introduced in the enhanced feature extraction network by changing the original concat to BiFPN_concat, adding learnable weights to each input feature to learn the importance of different input features, distinguishing the importance of different features in the feature fusion process, and better focusing on the target information to be detected. The experimental results show that the mAP of the improved algorithm is 92.65%, which is an average accuracy improvement of 2.55% over the original YOLOX-S algorithm and meets the practical requirements.