Research On Mask-occluded Face Image Recognition Algorithm For Smart Tool Cabine

In the past thirty years,due to the rapid development of computer technology,face recognition has gradually differentiated from traditional image processing and has become a hot topic today.The face recognition is to detect and position the face of the image or video,and identify the features of the face from it to recognize identification.According to the current epidemic prevention,the face recognition system of the factory intelligent tool cabinet need to recognize the factory operators who wear mask.Based on the deep learning algorithm,the improvement and optimization detection and identification algorithm make it accuracy and speed to apply requirements about detection of the face with mask.The main work contents of this article are as follows:Firstly,the data set is created for training and preprocessing.For face detection and identification of masks wearing,current public data sets which can be used is not enough.So with the tool called Dlib,public data sets can be transformed to sets of face with mask.Then,the new sets get labelled.Since the quality of the image has a great influence on the speed of the training and the accuracy of it,the image set gets denoised,cut,and augmented.Secondly,based on the homemade face data set,the traditional Haar-Like + Ada Boost algorithm and the current popular YOLOv3 face detection algorithm are compared on multimask face images.These two algorithms are compared by qualifying the specific test chart.Although the YOLOv3 detection algorithm leads ahead to the Haar-Like + Adaboost algorithm in the detection accuracy and speed,there is a false detection and missed detection when detecting a small multi-mask face,it has a great potential to be improved.Thirdly,there is a problem of detecting small multi-mask faces in complex scenes,so a face with mask detection algorithm based on improved YOLOv3 is proposed.First,the original Dark Net-53 is replaced with the CBRes Net50-DCN as the backbone networks,where the DCN convolution enhances the robustness of the model to the slight deformation.The attention mechanism of mixed domain channel is used to enhance the focus of key features.Then,the contact between the layers is improved by Bidirectional Feature Pyramid Network(Bi FPN).At the same time,DIo U is introduced as a positioning loss function,further enhances the detection accuracy of the model.Then,based on the improved detection algorithm and the improved face recognition algorithm Face Net,the face recognition of different faces is accommodated by the multi-model of the different dimensions to maximize the current facial features.Experiments show that the improved mask face detection algorithm has superior detection performance on existing image data set,but the detection speed has decreased.Meanwhile the improved recognition algorithm is more flexibly utilized for the acquired feature information,and it can achieve a higher mask face identification effect.Fourthly,the reasoning speed of the mask face detection is enhanced by building a lightweight network and pruning network parameters.First,a lightweight face detection network is built by introducing a lightweight convolution structure,then compares with the method of removing the redundant connection of the network using the network pruning.Experiments show that the mask face detection algorithm with lightweight structure can provide more efficient real-time,and keep high accuracy.Finally,based on the face detection and recognition algorithm of the mask face,the face recognition system of industrial smart tool cabinet is designed based on Qt.The system uses Qt combined with the My SQL to build the main program interface to cover the face detection and recognition algorithm,and it can perform increasing,deletion,modification,insertion of face identity information at the front end of the information processing,basically realizing the function of face recognition.