Mask recognition scheme based on improved Vision Transformer

At this stage, extracting effective mask features under more cumbersome feature information has become the focus of research on mask recognition. Considering the fact that the traditional ViT (Vision Transformer) cannot obtain all mask features due to the fixed down sampling rate, a mask recognition scheme based on improved Vision Transformer is proposed in this paper. On account of ViT network, the programme introduces DNN module in OpenCV to eliminate the interference of non face factors. In the same time, with the combination of Dlib module and ViT model, the practical application model is improved to enhance the learning and processing ability of the algorithm for detailed information. The experimental results show that the map of this scheme reaches 97%, overall_ The accuracy reaches 99%, which is improved by 3.3% and 1.1% respectively compared with the original ViT algorithm. The improved scheme has good accuracy and practicability for mask wearing recognition, and can effectively meet the actual needs of epidemic prevention and control.