Biometric identification technology is the use of human physiological characteristics or behavioral characteristics,to carry out personal identity identification.The retina is an extremely fixed biological feature which can not be forged because it is "hidden" and non-contact,not easy to wear and aging,Therefore,in the Internet era,the identification of human retina features can be used instead of password to ensure information security.In this thesis,convolutional neural network(CNN)is used to extract retinal features,and long and short-term memory Network(LSTM)classification model is used to study biometric recognition system based on human retinal images.The main research work of this paper includes the following points:(1)Sharpening and rotation are used to enhance the image data set to improve the contrast and resolution.(2)The convolutional neural network(CNN)was used to extract retinal features,and the long and short-term memory Network(LSTM)classification model was used for biometric recognition.(3)The Epoch(the number of images to be recognized),the Units parameters of LSTM,batch-size(sample set size)and iteration times were optimized,and the simulation comparison of model training was conducted by using MATLAB simulation tool.The simulation results show that the recognition accuracy of the same model is different in different age groups,and most age groups model 4 has higher recognition accuracy than the previous three models. |