Non-contact Palm Feature Recognition Based On Convolutional Neural Network

In the society where information exchanges are increasingly frequent,the demand for personal identity authentication is increasing.Biometric identification has higher reliability than traditional identity authentication methods.The palm of the hand contains rich vein information and palmprint information.Combined with the information of veins and palm prints,it can better represent the characteristics of the palm.It is used more and more in the access control system.The contact and non-contact acquisition methods are the two main acquisition methods for palm images.This article uses the palm image database acquired under the contact method of the Hong Kong Polytechnic University and the palm image database acquired under the non-contact method of the Chinese Academy of Sciences as experimental data.The neural network structure is studied and optimized to make it suitable for palm image recognition.In this paper,an optimized convolutional neural network is studied.Palm parameters are identified by adjusting its parameters.Compared to traditional palm feature recognition algorithms,convolutional neural networks have feature learning capabilities and eliminate the need for preprocessing.This article is based on the LeNet-5 network model to optimize,adjust the network structure and parameters to realize the image recognition and classification of the palm of the hand,and then verify its performance based on the open palm image library,and compared with the traditional feature recognition algorithm.The results show that the optimized network model can effectively identify and classify the palm features.Although the convolutional neural network has a strong feature representation capability,it still requires a large number of training samples to train the network structure and parameters,but it also brings a large amount of calculation.For the problem of how to train a convolutional neural network on a database with a small amount of data,this paper studies a dual-channel network structure and applies it to palm recognition.The dual-channel network structure utilizes three convolutional layer palms to learn,and is finally classified by the SVM classifier.Verified on the open palm image database,the results show that the network structure has strong learning ability.Even in the absence of pretreatment,effective learning can be performed and adapted to individual differences.