With the rapid development of information technology,how to quickly and effectively improve the security of citizens’ identity information has gradually become a research hotspot.There are many inconveniences in the actual use of traditional identification technology.For example,keys,passwords,etc.,all have the risk of being forgotten or lost,and it is difficult to meet the needs of contemporary people for higher security,reliability and convenience.Recognition using human body features has gradually become a hot spot for identity identification,such as face recognition and fingerprint recognition.Palm vein recognition technology is one of human biometric recognition technologies.It has become a research hotspot at home and abroad because of its advantages of living body authentication,difficult forgery,and convenient use.However,the robustness of the existing palm vein feature extraction algorithms is not high,and the quality of palm vein images is relatively high.This paper mainly adopts the method of deep learning,proposes the optimization method of this paper on the existing model,and proposes improvement ideas for the important steps in the palm vein recognition process.It includes key links such as palm vein ROI acquisition method,palm key point location method,and palm vein image feature extraction method,and has achieved certain results.The main contributions and innovations of this paper are as follows: First,in view of the low robustness of the commonly used palm vein image region of interest interception algorithm,this paper designs a two-stage network for palm region of interest interception based on key point positioning,which is divided into Global localization network and local localization network.In order to improve the positioning accuracy of key points,the existing Huber loss function is improved,which effectively improves the accuracy of key positioning.Second,in view of the fact that there are few samples in the palm vein standard database and the overfitting phenomenon that may occur in training the convolutional neural network,this paper proposes a data enhancement method,which effectively expands the database and provides support for the subsequent training of the convolutional neural network.Third,the traditional palm vein recognition algorithm has the problems of robustness and low recognition rate for palm vein image feature extraction.This paper uses the method of convolutional neural network to modify the existing compressed excitation attention module,and proposes an adaptive multi-layer shared perceptron channel attention module enhanced by pooling layers.By integrating the existing basic network model with the improved attention module,a convolutional neural network more suitable for the palm vein recognition task is designed.Finally,after a large number of comparative experiments,including the comparison of different feature vector dimensions and different network structures,and compared with the palm vein recognition algorithms in other papers,it is proved that the network model proposed in this paper is of great help for palm vein feature extraction.Fourth,integrate the key steps in the palm vein feature identification technology,design the palm vein identification system,and develop software to test the performance of the palm vein identification system. |