基于轻量化YOLOv5算法的口罩检测方法研究

Research on Mask Detection Method Based on Lightweight YOLOv5

由于疫情防控需要,人们在生活中都需要佩戴口罩,因此一种智能化口罩检测方法显的尤为重要。针对当前CNN网络和YOLOv5模型结构复杂、参数量大等特点,设计了一种轻量化YOLOv5s模型。将宽度系数由0.5变为0.4,采用GhostBottleneck替换原结构的C3模块,采用DSConv替换除输入端卷积层外的其余卷积层。实验结果表明,该模型将浮点数压缩为原来的四分之一,大小压缩为原模型的37.5%,GPU上训练时间缩短20%,提高效率的同时节省了空间大小,为在资源有限的终端上部署提供了可能。

Due to the needs of epidemic prevention and control,people need to wear masks in their lives,so an intelligent mask detection method is particularly important.Aiming at the cofqmplex structure and large number of parameters of the current CNN network and YOLOv5 model,a lightweight YOLOv5s model is designed.The width factor was changed from 0.5 to 0.4,and the original C3 module was replaced by GhostBottleneck,and the remaining convolutional layers except the input convolutional layer were replaced by DSConv.Experimental results show that the model compresses the floating-point number to 2.5 times the original,the size is compressed to 2.7 times,and the training time on the GPU is shortened by 20%,which improves efficiency and saves space,providing a possibility for deployment on terminals with limited resources.