A Safety Helmet Detection Method Based on Multi-Scale Feature Fusion and Improved Convolutional Structures

In industrial settings, monitoring the wearing of safety helmets by workers is key to ensuring job safety. However, traditional monitoring methods are difficult to meet real-time and high-precision requirements in complex environments. This article proposes an efficient helmet detection method based on the YOLOv8 deep learning model, and makes the following improvements for complex scenes: 1) Introducing a multi-scale feature fusion mechanism to optimize the detection performance of small targets (such as long-distance helmets) and large targets; 2) Adopting an improved convolution structure that combines depthwise separable convolution and dynamic convolution to enhance computational efficiency and model expressiveness. The experimental results show that the mAP of this scheme reaches 92.30/0 in complex backgrounds, occlusions, and high-density scenes, and the inference speed reaches 35 frames per second, which is significantly better than traditional methods. The research provides an efficient and reliable solution for industrial safety monitoring.