Safety Helmet Detection Method Based on Improved YOLOv5
In view of the common problems of existing safety helmet detection methods in complex scenes, such as poor detection effect of small targets, prone to misdetection and omission, low robustness,etc., a safety helmet de-tection method based on improved YOLOv5 was proposed. By adding SimAM attention mechanism to the backbone network, the model can characterize and strengthen the different importance of 3D feature points without adding ad-ditional parameters. A small target detection layer is added to the neck network to enrich the fine-grained informa-tion of the target. and Decoupled-Head is used instead of the YOLOHead module of the original model to carry out the classification and regression tasks separately. The experimental results show that the average accuracy of this method is 93.17％, which can meet the requirements of safety helmet detection in complex scenes.
YOLOv5safety helmet detectionsmall targetattention mechanismfine grained information
万方数据
维普
