Enhanced YOLOX-L Algorithm with CBAM for Construction Safety Detection

Abstract

Many safety accidents in the construction industry are attributed to unsafe behavior, particularly the incorrect use of safety protective equipment. To address this issue, this paper proposes an enhanced YOLOX-L algorithm, which integrates the Convolutional Block Attention Module (CBAM) to improve the Neck component of the original network. This modification enhances the network's ability to extract features at various scales. While maintaining real-time detection speed, the mAP (mean Average Precision) trained with a custom dataset reaches 85.96%, which is approximately 1.1% higher than the original YOLOX-L algorithm. This effectively enables fast and accurate detection of unsafe behavior.