Research On Mask Face Detection Based On Improved YOLOv5

In recent years,novel coronavirus has broken out violently all over the world,and masks have become a necessary protective tool for our daily travel.However,in the case of wearing a mask,the face part is covered by a large area,which makes the face detection technology a great challenge.Under the condition of face occlusion,effective face detection has become the primary problem to be solved in target detection,and is also a hot issue in the current scientific research field.YOLOv5 is now a common model for target detection based on deep learning,so this thesis improves the detection of face mask based on YOLOv5.The main innovations and work are as follows:(1)A DCS＿YOLO model based on CBAM attention mechanism for mask face detection is proposed.1,First of all,in view of the low detection accuracy of YOLOv5 n for the mask face,the CBAM attention mechanism is introduced in the Neck part of the model.The input mask face information focuses on the information that is more critical to the current monitoring task,reducing the attention to other information,which improves the detection accuracy of the mask face.2,In view of the slight increase in the calculation amount of the model after the introduction of CBAM,the ordinary convolution of the Neck part is replaced by the deep separable convolution,which reduces the calculation pressure of the model.3,The activation function in the network is changed to Hardswish activation function,which improves the accuracy of the model again without increasing the amount of computation.The improved DCS＿YOLO network model has further improved the detection accuracy of the mask face dataset.In DCS＿YOLO,D refers to separable convolution of depth,C refers to CBAM attention mechanism,and S refers to Hardswish activation function.(2)Based on the DCS＿YOLO model,the lightweight MASK＿YOLO face detection network model is proposed.1,First of all,the Mobile Netv3 small network is used to replace the network of the DCS＿YOLO model Backone,and the computing load of the computer on the model has been significantly reduced.2,To solve the problem that the mask face dataset is not rich in samples and the model convergence speed is slow,Mosaic data enhancement strategy is adopted,which also improves the accuracy of the model.3,In order to solve the problem of inaccurate regression of facial masks,a new loss function,P-EIo U loss function,was obtained based on the improvement of EIo U loss function.Finally,an improved MASK＿YOLO network model is obtained.The performance of the model is tested through ablation experiment,contrast experiment and robustness experiment,and the comparison between various lightweight network models further proves the excellent performance of the MASK＿YOLO mask face detection model.