Research And Application Of Small Sample Iris Recognition Technology Based On Deep Learning

With the development of the information age,people's privacy and information have become more insecure,and the importance of identification is becoming increasingly prominent.Due to its uniqueness,security,reliability and convenience,biometrics technology has gradually replaced the traditional identity authentication method.Compared with other biometrics,iris is currently favored by the field of identity security because of its unmatched advantages.The field of biometrics has become the most worthy of development and attention.Due to the rapid development of deep learning technology in recent years,with its excellent performance in the field of image classification,image classification with the help of deep learning technology has gradually replaced the traditional image classification based on manual features,but deep learning methods usually require massive Training data to accurately classify specific image categories.Therefore,studying how to apply deep learning methods to small sample iris image classification has become a hot topic in the field of iris recognition.?1?An iris recognition algorithm based on convolutional neural network transfer learning is proposed to solve the problem that the deep learning method is not effective in small sample iris data sets.This algorithm is based on the idea of transfer learning,based on the pre-training network VGG16,to carry out model migration,and build an ICP-VGG model using the iris recognition task.First,the convolutional base of the pre-trained network VGG16 and its weight parameters are retained;secondly,a custom dense connection classifier is added after the convolutional base.At the same time,in order to reduce the over-fitting phenomenon of the ICP-VGG model on the small sample iris data set,the following actions are taken:using data enhancement technology to expand the iris data set sample;adding a Dropout layer to the custom dense connection classifier;introducing₂ regularization The term improves the cross-entropy loss function;after the model is successfully constructed,multiple sets of experiments are used to determine the optimal parameters of the ICP-VGG model.Finally,through comparison experiments,it can be verified that this algorithm can well apply deep learning methods to small sample iris data sets.?2?An iris edge detection algorithm based on adaptive Canny operator and multi-directional Sobel operator is proposed to optimize the processing of iris images,further improve the application effect of the ICP-VGG model proposed in this paper on small sample iris data sets,and improve Iris recognition accuracy.The algorithm first uses median filtering to preprocess the iris image,and secondly uses fourth-order moments containing standard deviations to count the distribution of pixels in different iris images and determine the spatial scale coefficients and template window size in the Gaussian filter.Then combined with Sobel operator,the gradient operation is extended to 4 directions,and weighted to represent the gradient information of the entire iris image.Finally,the Otsu algorithm is used to select the high and low thresholds.Experimental results show that the proposed algorithm can better perform edge detection on the iris image,and the improved iris image processing method can further improve the recognition accuracy of the ICP-VGG model in the small sample iris data set.?3?An iris recognition prototype system is designed and developed.Based on the above two algorithms,the two algorithms are packaged into an image preprocessing module and a model training module and applied to the system.It is verified that this paper is based on deep learning Availability of a Small Sample Iris Recognition Algorithm to Develop a Prototype System.