Improved method based on complex background YOLOv8 optimization

Establishing a lightweight yet high-precision object detection algorithm is essential for accurately assessing the helmet-wearing status of workers in complex industrial environments. Helmet detection poses significant challenges due to several factors, including small target size, intricate backgrounds, and the need to balance model compactness with detection accuracy. In this article, we propose an improved method based on the YOLOv8 model. Our approach enhances the model’s capability to detect small targets within complex settings. Additionally, we introduce a novel module that integrates Depthwise Convolution (DWConv) with the RepVGG network structure into the backbone feature extraction network. This integration aims to reduce model complexity while improving accuracy. To further enhance helmet detection performance, we incorporated a C2fDFC attention mechanism module into the feature extraction network. This addition reduces both parameter count and computational load, ultimately leading to superior helmet detection outcomes. We used the Hard Hat Workers Object Detection Dataset released by Roboflow, which is a dataset of hard hats in complex situations, to verify our research method. Our proposed model achieved an optimal mean Average Precision at IoU 0.50 (mAP50) of 66.1 %. These results demonstrate the effectiveness of our method in achieving lightweight deployment alongside high-precision helmet detection.