Research on Finger Vein Recognition Based on Sub-Convolutional Neural Network

Abstract

Aiming at the problem that the traditional deep learning finger vein recognition algorithm reduces the recognition rate when there are fewer training samples and training times, a finger vein recognition method based on multi-convolution and multi-scale features is proposed. Two relatively independent sub-convolutional networks with different granularities are used for feature extraction. The convolutional layer uses the LeakyReLU activation function, and the pooling layer uses the maximum value downsampling method. For the object classification of the model, feature extraction from a single layer of different feature extraction as a result, the samples are finally trained through the softmax classifier. Experimental results show that the use of two relatively independent sub-convolutional networks with different granularities can more effectively extract features, and the accuracy rate can reach 95.1% when the sample set is small. At the same time, in the case of picture rotation, the error rate of this method is only 0.0373.