Research On Face Recognition Based On Adaptation Layer Network

Face recognition is one of the most successful directions in business for deep learning,in the area of computer vision,face recognition has always been a popular research direction.Today face recognition is widely used,for example,payment,attendance,justice,factory,education and many more.Face recognition is to find the faces that need to be identified in the videos or images.The face recognition task is divided into two phases,the first phase is detection the second phase is identification.At present,scholars at home and abroad have pay extensively attention to face recognition.Several mature deep learning models for face recognition today include FaceNet,OpenFace,facerecognition,etc.But,at the training process,the types of characters considered in these model training processes are relatively fixed,the face recognition model itself will add restrictions during training,and it is not easy to generalize.Therefore,for how to apply these classic face recognition models to other environments,especially in the absence of training data or the lack of data labels.There are still great challenges in making these classic models into face recognition systems of different environments and different races.This article will use the method of transfer learning and the method of adaptation layer to train the classic network model with the source data set and the target data set.By reducing the distribution difference between the source data set and the target data set to improve precision effect on the target dataset in the classic face recognition model.Main work is as follows:? At first,Based on the existing research theory of the adaptation layer and the current network structure of classic face recognition networks,analyzing the feasibility and practicability of the combination of the two,this paper proposes a dual-channel adaptation layer face recognition network structure.? Then,We are focus on the classic FaceNet network model in the existing face recognition task.Then,analyzing DDC(Deep Domain Confusion)network model.and getting the method for changing into the dual-channel training model in theory.? Finally,analyzing The training strategy of the adaptation layer network is obtained through the DDC network,and the FaceNet network model is used to implement the establishment of a face recognition network model based on the adaptation layer.By retraining the face recognition pre-training model to reduce the distribution difference between the source dataset and the target dataset,finally,it can achieve the goal of improving the accuracy of the target dataset on the face recognition network.The article makes an in-depth analysis of the network's selection of adaptation layer parameters and loss functions.Compared with the classical face recognition algorithms,the experimental results show that the face recognition algorithm based on the adaptation layer is more effective.