Finger Vein Recognition Algorithm Based On Improved Residual Network

With the advancement of biometric recognition technology,these technologies have gained broad application prospects.Finger vein recognition technology is more firm,efficient and stable when paralleled with other biometric recognition technologies.However,.the currently acquired finger vein images generally have problems such as low image quality and the degree of freedom caused by the finger's degree of freedom,which poses great challenges to finger vein recognition technology.Traditional finger vein recognition technology extracts features based on image texture,feature points and other details.If the image quality is poor,the extracted finger vein features will be unstable,which will reduce the accuracy of finger vein recognition.In view of the above problems,this paper introduce a finger vein recognition algorithm on the basis of improved residual network.In the deep learning algorithm,there are fewer data sets of the finger veins,how to use the limited data sets to train to obtain a robust model performance is one of the main problems.The distinguishing feature of the finger vein recognition technology is the details of the finger vein picture.The network needs to be redesigned to minimize the loss of information in the feature map and extract the robust finger vein features.At present,common convolutional neural networks are used for classification purposes.For finger vein recognition technology,it is impractical to collect pictures of finger veins of all categories in advance.Therefore,it is necessary to redesign the network's supervised signals so that the network learns.To the characteristics of finger veins with discriminative performance,the main tasks of this paper is as follows:?1?Aiming at the problem of finger offset during the acquisition of finger veins,this paper proposes a preprocessing method for angle correction and extraction of the region of interest,which can eliminate the interference caused by the offset and obtain a normalization region of interest after the transformation.?2?Aiming at the problem that there are few open finger vein data sets,this paper uses traditional image processing algorithms to amplify the data sets,and uses pre-training weights for the training phase of the network to alleviate the lack of finger vein data sets..?3?In order to reduce the information loss of the feature map during the network training process,this paper uses a convolutional neural network with an improved residual module to reduce the information loss caused by the large step size of the small convolution kernel.The kernel is replaced by multiple small convolution kernels,which reduces the network parameters and improves the expression ability of the model.?4?In order to further improve the performance of the model,the residual network uses an improved activation function to improve the model's ability to express.At the same time,in order to improve the model's ability to discriminate unknown categories,the network uses Softmax Loss to increase the distance between classes of the finger vein features.Center Loss is used to reduce the intra-class distance,and the two are used as the loss function of the network to constrain the back propagation process of the network.In order to verify the performance of the finger vein recognition algorithm used in this paper,multiple experimental comparisons on two public data sets of FV-USM and MMCBNU₆000 show that the improved residual network has improved the performance of the algorithm to a certain extent and the accuracy of the recognition has also gained improvement.