Finger Vein Recognition by Channel-Wise Feature

In order to improve the ability of convolutional neural networks to recognize finger veins, a MobileNet V2 network combining the importance of channels is proposed. Specifically, for the features extracted by MobileNet2, channel importance analysis is performed, and unimportant feature channels are compressed to improve network feature representation capability. At the same time, the network model by using triplet loss can be extracted with the characteristics of inter-class distribution and intra-class compactness, which improves the discriminant ability of the network model. Experiments are carried out on the MultiView-FV vein dataset, and the results show that the proposed method is efficient and effective.