The Behavior Detection Method for Live Working Personnel Based on an Improved YOLOv4 Lightweight Network

To meet the dual requirements of real-time performance and high accuracy in live-line workers’ behavior detection tasks, a detection method based on an improved YOLOv4 lightweight network is proposed. Firstly, the method replaces the CSPDarknet53 backbone of YOLOv4 with the GhostNet residual structure and optimizes the convolutional neural network using the Ghost module, significantly reducing computation and parameter overhead. Secondly, the introduction of the Global Attention Module (GAM) enhances the extraction of key features, improving detection accuracy. Finally, the Focal Loss is employed to optimize the loss function of the YOLOv4 network model, further enhancing detection performance and reducing computational resource consumption. Experimental results show that compared with the traditional YOLOv4, the proposed model reduces the parameter count by 10.3 M, improves mAP by $3.55 \%$, and increases detection speed by 1.12 times. The model demonstrates significant advantages in comprehensive detection performance, efficiently achieving real-time detection of live-line workers’ behavior, and provides strong technical support for the safe operation and maintenance of transmission line tasks.