Attention-based Deep Learning Model Using Adaptive Margin Loss For Finger-Vein Recognition

As the importance of personal and private information continues to increase, biometrics recognition using deep learning (DL) has been attracting more attention. However, traditional finger-vein recognition models have struggled to adapt to the varying quality of images used in everyday life, leading to questions about the model's effectiveness. Because of this, this study proposes the implementation of the CoAtNet model and the adaptive margin loss function (AML) to achieve finger-vein recognition. This method specifically adjusts the difficulty of the training based on different image qualities. The experimental results indicate that the proposed model using AML can have an accuracy rate of 99.80 % in the public dataset FV-USM.