FEFD-YOLOV5: A Helmet Detection Algorithm Combined with Feature Enhancement and Feature Denoising

In intelligent surveillance of construction sites, safety helmet detection is great significance. However, due to the small size helmets and presence high levels noise in scenarios, existing methods often encounter issues related insufficient accuracy robustness. To address this challenge, paper introduces a new algorithm, FEFD-YOLOV5. The FEFD-YOLOV5 algorithm enhances performance by adding shallow head specifically for target incorporating an SENet channel attention module compress global spatial information, thus improving model’s mean average precision (mAP) corresponding scenarios. Additionally, novel denoise introduced ensuring model maintains robustness under various conditions, thereby enhancing generalization capability meet real-world scenario demands. Experimental results show that proposed improved achieves 94.89% noiseless environments, still reaches 91.55% high-noise demonstrating superior efficacy compared original algorithm.