Research on a Method for Identifying the Safety Behavior of Construction Site Personnel Based on Lightweight Yolov8

This research paper tackles the problems of ineffectiveness in conventional regulatory approaches for identifying safety behaviors of construction workers and the inability of existing deep learning models to meet real-time detection requirements. It proposes an improved model based on a modified version of YOLOv8. The model replaces the backbone network of YOLOv8 with MobileNetV3 and introduces the CA attention mechanism. Experiments were conducted using a custom-built dataset of 1,206 images of safety helmets and safety vestss (with five category labels), and ablation experiments were used to validate the CA attention mechanism. The findings indicate that when compared to the initial YOLOv8n, the size of the enhanced YOLOv8 - MobileNetV3 - CA model has been decreased to 4.5 mm. model has been reduced in size to $4.9 \text{MB}(80.3 \%$ of the original model), with a 29.6% reduction in computational load (Flops), Simultaneously, it attains an accuracy ratio of 91.7% and a recall ratio of 90.0%. This achieves model lightweighting and improved detection speed while maintaining detection accuracy, meeting the real-time monitoring requirements of construction sites and providing a feasible solution to the problem. This achieves model lightweighting and improved detection speed while maintaining detection accuracy, meeting the real-time monitoring requirements of construction sites and providing a feasible solution for intelligent safety supervision in construction.