Safety Helmet Detection in Electrical Power Scenes based on Improved Lightweight YOLOv8

Abstract

Abstract To tackle the challenges of diverse targets, complex scenes, and partial occlusion in safety management during electrical field operations, the YOLO series algorithm, recognized for its exceptional accuracy and swift processing capabilities, has been applied to various scene detection tasks. To ascertain if workers have donned safety helmets and ensure the safety of electrical field operations, we propose a lightweight algorithm based on the improved YOLOv8 for constructing a digital safety helmet detection system. By incorporating the VoV-GSCSP module, we reduced model complexity, decreased computational load, and improved detection accuracy. Simultaneously, by combining the GSConv module, we enhanced the network’s feature extraction capability, enabling the network to adapt more rapidly and accurately to various complex electrical scenes, thereby strengthening the network’s robustness in safety helmet detection. Finally, we validated the effectiveness of the proposed model using the pre-existing dataset for safety helmet detection.