Research On Palm Vein Feature Extraction And Recognition Algorithm Based On Deep Learning

With the development of economy and technology,people pay more and more attention to the security of personal information.At the same time,it has become an important aspect of social management to strengthen the research of personal identification technology and improve the information security management.As one of the biometric recognition,palm vein recognition has unique characteristics such as uniqueness,difficulty in replication,living identification,and high security level,so it has attracted more and more attention from researchers and industry.However,there are two problems in palm vein recognition: On the one hand,because of many factors,the acquired vein image contains not only vein but also noise and irregular shadow,which makes it difficult to extract the distinguishing information of vein features effective ly,thus reducing the recognition accuracy;On the other hand,in many application scenarios,due to insufficient storage capacity of the recognition system or to protect user's privacy,each user in the training set has only one sample.However,the existing vein recognit io n models are often prone to overfit on such a small sample database or are difficult to be ful y trained,so the recognition accuracy is greatly reduced.Therefore,the recognit io n based on a single image is still a chal enging problem.Meanwhile,deep learning has made remarkable achievements in computer vision,natural language processing,and speech recognition.In this case,we apply deep learning to the feature extraction and recognition of palm vein,and carries out the following research.We introduce the convolutional neural network and generative adversarial network.Firstly,we introduce the basic components of convolutional neural network and a classic convolutional neural network-LeNet5.Secondly,we introduce the loss function and training process of the generative adversarial network in detail.Lastly,the deep convolutional generative adversarial network is introduced by combining the convolutional neural network and generative adversarial network.We study the feature extraction of palm vein,and propose a palm vein segmentat io n algorithm based on U-shaped generative adversarial network.First of al,this method extracts the region of interest of the palm vein through image preprocessing techniques such as image binarization,edge detection,and correction.Next,we label palm vein images by integrating multiple segmentation methods.Once more,a U-shaped generative adversarial network is constructed,which means the U-shaped network as the generative network and the convolutional neural network as the adversarial network.Finally,through the alternating training of the two networks,we learn the internal distribution of the palm vein and achieve the segmentation of palm vein image.The obtained equal error rate is 0.33% and 0.026% separately on the CASIA database and PolyU database.The experiment results show that the proposed algorithm is superior to the existing segmentation methods.We study the palm vein recognition of a single training sample,and propose a recognition algorithm based on multi-scale generative adversarial network.Firstly,the real distribution of a single training image are learned through the multi-scale generative adversarial network(SinGAN),and the training set is established by generating independent and identically distributed palm vein images.Once more,the convolutio na l neural network recognition model is established,and the training set is used to train it.Finally,we explore the palm vein recognition accuracy and robustness among the different numbers.The experiment results show that the SinGAN network has a good performance of generating samples,and improve effectively the recognition accuracy and robustness.