Mask Wearing Recognition Based on Fusion Algorithm

Abstract

In view of the low accuracy of mask wearing recognition caused by factors such as multiple and small target recognition, extreme lighting conditions and low definition, this paper proposes a fusion algorithm that integrates the attention mechanism FcaNet and dense convolution blocks into YOLOv5s. The fusion algorithm is used to address the issue including small target identification and low-resolution facial image recognition. This paper obtains data from open-source datasets RMFD (Real-World Masked Face Dataset) and celebA, uses fusion algorithm for feature extraction, and finally uses FaceNet and support vector machine (SVM) for face mask wearing recognition. After experimental comparison, the fusion algorithm can achieve 97.55% accuracy of face mask wearing recognition under different complex environments, and the average frame rate is 30.3FPS. The algorithm model improves the recognition rate of mask wearing under occlusion conditions and with different clarity, has higher recognition performance, and can be used for mask wearing recognition when entering public places.