Research On Face Recognition Machine Learning Method Based On Mobile Terminal

With the rapid development of the mobile Internet and the popularization of smart mobile terminals,more and more users are gradually switching from the traditional PC-side Internet access to the mobile Internet based on mobile smart terminals.All kinds of application software,while subverting the user's living habits,involve the user's personal privacy,which inevitably brings the threat of security leakage of mobile information.Therefore,information security based on mobile terminals has become a popular research direction.Face recognition technology is mainly based on human facial feature information for identity recognition.It integrates artificial intelligence,machine learning,image processing and many other technologies,and its recognition accuracy rate has gradually improved.Especially in recent years,the development of artificial intelligence technology has pushed face recognition technology to a stage of accelerated development.At present,face recognition and fingerprint recognition have become the main identity authentication of mobile terminals.However,due to the limited computing and storage resources of mobile terminals,machine learning related technologies are difficult to use on mobile terminals.This paper mainly studies the realization of the face recognition machine learning method on the mobile terminal,and uses the mechanism of separating the training model on the PC side and the interface development on the mobile side.Nowadays,deep learning technology is widely used in the field of face recognition,but its model parameter scale is large and the algorithm complexity is high,it is difficult to deploy and use on the mobile terminal.This paper first designs the face preprocessing algorithm,filters the background area in the image through face preprocessing,and then extracts the face features through the convolutional neural network,performs face recognition training,and finally performs on the mobile terminal Interface development,complete real-time face recognition function.In terms of face preprocessing,this paper adopts the multi-task learning method and designs three cascading methods of convolutional neural networks in model construction.Combining the three tasks of whether the image has a human face,detecting the face area,and locating the key points of the face,the three loss functions are simultaneously optimized during the training process.In the construction of the network model,three convolutional neural network cascades are used to filter the face area in the image through multiple networks,which improves the accuracy of recognition.By this method,the background in the image can be filtered,and finally only the face area and less background are left,which can improve the accuracy of subsequent face recognition.In this paper,we design a lightweight convolutional neural network to extract image features,and give three algorithms: deep convolution,grouping convolution,and channel rearrangement to reduce the parameter scale and calculation complexity.In this paper,Triplet loss is designed as a loss function to optimize the network and train a model for face recognition.Through the verification on the test set,the network has good results in accuracy and detection efficiency.Finally,the model of the convolutional neural network is deployed on the mobile terminal.The model training and real-time detection are separated.The model training is performed on the PC and the mobile terminal performs real-time face recognition.The entire mobile terminal development can be divided into three modules: image acquisition and preprocessing,model deployment and forward propagation,and data storage and retrieval.The mobile terminal extracts the real-time data of the preview frame of the camera,extracts the facial features through the convolutional neural network,and finally completes the task of facial recognition.The whole system can detect and recognize human faces in real time and complete the realtime face recognition function.