Small Sample Face Recognition Method and System Design Based on Improved Siamese Network
In this paper,a face recognition method based on improved twin network is proposed to solve the problem of small sample recognition in view of the situation that the data set of face recognition in universities and other large organizations is multiple classes and small samples.By referring to the application of original twin network in different small sample projects,a new improved twin network is constructed by adding SE attention mechanism to feature extraction,using Mahalanobis distance to optimize distance function and classifier,and the corresponding face recognition system is designed.After the training on the AT&T data set,the improved network model achieved 99.55%accuracy in the corresponding test set.Compared with the original twin network and the traditional model framework such as ResNet18 and VGG16,the model accuracy was increased by 1 to 5 percentage points respectively.Experiments show that the proposed method has high accuracy and robustness.
万方数据
维普
