Research On Iris Recognition Method Based On Two-stream Convolutional Neural Network

In today's information age,a large amount of information is stored in the form of data in personal or enterprise hardware equipment.What technology is used to protect information has become a core topic.Iris recognition has high uniqueness,strong stability and non-invasiveness,which makes it one of the safest and most stable recognition technologies.It bears the responsibility of protecting information,Therefore,it has become one of the important research fields.This paper applies convolutional neural networks to iris recognition,performs classification tasks on iris images,and is used for the identification of people.It also addresses the problems of overfitting and recognition instability in traditional iris recognition research by traditional convolutional neural networks.Corresponding improvements have been made.Experiments show that the improved convolutional neural network model has higher recognition accuracy and stronger robustness in iris recognition.The specific methods and improvement ideas are as follows:(1)Aiming at the overfitting problem caused by the small amount of iris data and single iris features,a convolutional neural network based iris recognition method based on transfer learning is proposed.First,the iris image is segmented and normalized to obtain regularity.Iris image,and then train the ImageNet dataset with classic convolutional neural network models(VGGNet,GoogleNet,ResNet)to obtain training parameters,and then migrate this parameter as an initial weight to the iris dataset(CASIA-Iris-Interval,In the training of CASIA-Iris-Lamp,CASIA-Iris-Thousand,and IITD Databas),the Softmax classifier was used to classify the iris,and a comparison experiment was performed with the network without transfer learning.The experiments show that the network based on transfer learning relieves Overfitting phenomenon,and improved the accuracy and stability of iris recognition.The average recognition accuracy of ResNet,which performed best in the iris data set,reached 97.7%,an increase of 3.4%,a range of 2.5%,and a decrease of 5.2.%,Realizing accurate and stable recognition in the iris data set;(2)In view of the lack of strong robustness of the network in the classification of iris images,based on GoogleNet 's high computing performance and ResNet 's high generalization performance,a dual-stream convolutional neural network model is proposed,and GoogleNet and ResNet are used to analyze iris data Set training,get two recognition results,and use the scoring mechanism fusion method proposed in this paper to fuse the classification confidence of the two networks,calculate the comprehensive classification accuracy value,and use the comprehensive classification accuracy value feedback to control the dual-stream convolutional neural network model.During the training process,determine the optimal scoring mechanism fusion method weight value and improve the average recognition accuracy of the network model.The experimental results show that the model further improves the accuracy of iris recognition based on the transfer learning model.The average recognition accuracy on the data set(CASIA-Iris-Interval,CASIA-Iris-Lamp,CASIA-Iris-Thousand and IITD Database)reached 98.4%,an increase of 0.7%,and improved the robustness of the network.After reconstructing the network structure using the above improved ideas,training and testing were performed on the iris data set.The experimental results show that the proposed migration model and dual-flow model have improved the accuracy of iris recognition and the stability of recognition to a certain extent.It is more robust and has certain reference significance,which promotes the application of deep learning in iris recognition technology.