Safety Helmet Wearing Detection Algorithm Based on Lightweight YOLOv5s
Wearing a safety helmet during construction is one of the good protective measures for various high-risk industries nowadays.In order to solve the problem of large model parameters that cannot be deployed on mobile and embedded devices,the YOLOv5-MN detection algorithm is proposed for helmet wearing detection.Firstly,the lightweight module GhostNet is introduced into the backbone network of YOLOv5 for optimization.The input feature map is divided into two parts and convolved to varying degrees to reduce computational complexity.Secondly,a new feature fusion network structure BiFPN is adopted to fuse feature maps at different levels to obtain richer semantic information.Finally,an ECA attention mechanism is added by introducing channel attention modules on the feature map to dynamically adjust the importance between channels,in order to enhance the model's perceptual ability.The experimental results show that the complexity of the lightweight YOLOv5 model is significantly reduced,and the inference speed is significantly improved.
NETL
NSTL
万方数据
