ResNet多输出特征的手指静脉识别

Finger Vein Recognition Based on ResNet Multi Output Feature

为了解决成像环境复杂、生物组织引起红外光散射导致识别准确率低的问题,在基于神经网络分类的手指静脉识别技术基础上,提出神经网络拓展结构模型,并采用机器学习方法计算区分性好的特征,弥补原神经网络引起的部分特征缺失,避免了模型迁移到小型数据库上由于样本数量不足带来的性能下降问题.利用ResNet18、ResNet34和ResNet50三种神经网络在手指静脉数据库上进行的分析与实验表明,拓展结构模型显著提高了识别性能.

In order to solve the problem of complex imaging environment and low recognition accuracy caused by infrared light scattering resulting in biological tissues,an extended structure model of neural network was proposed on the basis of finger vein recognition technology based on neural network classification,and machine learning method was used to calculate features with good discrimination for making up for part of the missing features caused by the original neural network.This avoids the perfor⁃mance degradation caused by model migration to a small database due to insufficient sample size.The analysis and experiment on finger vein database using ResNet18,ResNet34 and ResNet50 neural networks show that the extended structure model signifi⁃cantly improves the recognition performance.