 The study aims to improve safety helmet detection in complex scenarios by incorporating the ECA channel attention mechanism into the YOL-5 backbone network, adopting a weighted bi-directional feature pyramid network structure, by FPN, and introducing a decoupling head in YOL-5. The enhanced model achieved an average accuracy of 95.9% on a custom-made helmet dataset, a 3.0% point increase compared to the original YOL-5 model. This article was authored by Chunxian, Wang Mingliu, and YuYu.