The Research Of Face Recognition Based On Transfer Learning

At present,more and more applications require identification technology,while traditional identification technologies have disadvantages such as being inconvenient to carry,easily lost,damaged,and hacked.Therefore,face recognition in biometrics has attracted a lot of attention.Due to its distinctive features such as uniqueness,universality,operability and non-aggressiveness,the security is guaranteed.Therefore,face recognition applications are also more and more widely,such as security,military,civil and other fields.At the same time,the research of face recognition technology is also of great significance to the development of other image processing fields.In practical applications,the small sample data set is a very challenging issue for face recognition technology.The traditional face recognition method first extracts the features of the training sample data,and then performs classification and recognition.The small sample training data will lead to the traditional face recognition method is difficult to obtain a good recognition rate.The main idea of this thesis is how to classify face effectively and improve the recognition accuracy of face recognition with only a few face samples.Therefore,this paper presents a face recognition method based on transfer learning,the main work and innovation are:In view of the problems caused by the training data of the small sample,this paper presents a face recognition algorithm based on transfer learning convolution neural network,the main idea is based on the convolution neural network model,based on the model combines the transfer learning technology thought of.However,convolution neural network model prone to over-fitting in the training process,and the training time will be longer,for these two issues,this paper presents a shallow convolution neural network model,and in this Based on the shallow convolutional neural network,the idea of transfer learning technology is incorporated into the algorithm,and a convolution neural network algorithm based on transfer learning is proposed to improve the performance of the original convolutional neural network.First,a shallow convolution neural network model is designed,and parameters such as learning rate of convolutional neural network are tuned to ensure that the convolution neural network will not converge in a short time Second,by adding the idea of transfer learning algorithm,based on the designed shallow convolution neural network,a small amount ofnon-repetitive samples are retrained.Compared with the classification performance of convolution neural network,the new model Which can greatly improve the classification performance of the model under the premise of consuming a small amount of time and improve the robustness and recognition accuracy of the face recognition on the small sample training datasets.In order to verify the feasibility of this algorithm,two different face databases are selected to test and evaluate the proposed algorithm.Compared with the PCA+KNN,the SVM and the based convolutional neural network algorithm,the proposed algorithm in a short training time,The classification accuracy is greatly improved,which verifies the rationality of the algorithm.