Research And Implementation Of Face Recognition Based On Deep Learning

With the rapid development of machine learning and pattern recognition,image recognition technology combined with deep learning algorithms has been widely used.The face recognition algorithm based on deep learning overcomes the drawbacks of the traditional methods.The training model can be used to extract detailed facial information to achieve a higher recognition rate.This paper mainly studies the application of face recognition based on deep learning in real-time scenarios.Compared with traditional face recognition algorithms,the use of convolutional neural network models does not require people to design algorithms to extract features.In this paper,the collected face data is used to fine-tune the VGG-16 model based on the Caffe platform to obtain a new model.Finally,the face detection and recognition system in real-time scenarios was designed based on a new model of fine-tune.The main work of this paper is as follows:(1)Firstly,the background and status of face recognition technology are briefly introduced,and the usage method,working principle and advantage of deep learning tool Caffe,are described in detail.Then,it describes the gradient descending and back propagation algorithm in the traditional neural network.The theoretical basis of the convolutional neural network is introduced in detail,.and the key layers such as the convolutional layer and the pooled layer are elaborated.Finally,the network structure of the VGG-16 model to be used in this paper is analyzed.(2)Firstly,collect face data and preprocess the face data,the file name of each type of face picture is given a specific name,and it is converted into a data type that can be recognized by Caffe,and then by writing script files and constantly adjusting related parameters,a new model for collected face data is trained based on the VGG-16 model.Finally,through feature visualization and test results analysis,compared with the features extracted by traditional methods,the features extracted by the model can better express face information and have a better recognition rate.(3)Using the above-mentioned trained model,a camera-based face recognition software is designed on the Python platform,including three parts of camera image analysis,registration module and identification module.Then the system is tested under the real-time scenario,the results show that the software system has better recognition effect in the conditions of face blocking,facial expression changes and posture changes.In summary,this paper has completed the collection of face images and the establishment of a database.The recognition rate of the face recognition model through Finetune reached 97.03%.Through real-time face recognition verification of face image database,the designed face recognition software system can be applied to the realization of access control or time and attendance system based on face recognition.