Research On Contactless Palmprint Recognition Based On Deep Learning

With the development of Internet technology,palmprint recognition,as a biometric recognition technology,has gradually presented the characteristics of high accuracy,high security,and high user acceptance.The acquisition area of palmprint image is wide,the acquisition equipment requirements are low,and the acquisition method is easy to be accepted.However,the palmprint recognition algorithm is susceptible to noise,light intensity,and other factors.Due to the large area of palmprint features and rich texture information,it is difficult for palmprint recognition algorithm to effectively extract information representing palmprint features,and the utilization rate of texture information is not high,resulting in low recognition accuracy.Therefore,the study of palmprint recognition technology has important theoretical value and practical significance.In order to solve the above problems,this thesis studies the palmprint recognition method combining attention mechanism and residual network and optimizes the model.Since the contact collection of palm print requires users to touch the instrument and fix the position of the hand,the user friendliness is low,so this thesis adopts the non-contact collection of palm image.The friendly and hygienic non-contact collection of palm print is more acceptable to users.In order to make the experimental conditions closer to the real environment,based on the open palm print data set,this thesis constructs a palm print data set composed of palm images of different models of smart phones under the unrestricted background.In the stage of image preprocessing,the region of interest is extracted from open and self-built palmprint data sets,and the data is enhanced.By adding attention mechanism based on model Res Net50 and adopting a palmprint recognition method based on attention mechanism,the network model can identify the palmprint areas of interest more intensively,focus on the prominent features,allocate more weight,and ignore some features of non-key areas,which can improve the utilization rate of information.Thus,it can improve the accuracy of palmprint recognition.In order to further improve the performance and effect of the model,an improved residual network module was proposed on the basis of palmprint recognition based on attention mechanism.The model was optimized by constructing grouping convolution and combining with random SDPoint technology with adjustable training cost,loss function and optimization algorithm selection.Three evaluation indexes,accuracy,recall and ₁F,were used to evaluate the model.Experimental results show that the palmprint recognition method based on the improved residual network proposed in this thesis,The accuracy on the Poly U II dataset reached 98.69%,which is basically consistent with existing methods.The accuracy on the Tongji,IIT-D,and CASIA datasets reached 99.94%,99.32%,and 99.51%,which is an improvement of 0.05%,0.81%,and 0.27%compared to existing methods.The accuracy rate on the self-built palmprint dataset reached 98.37%.The experimental results demonstrate the effectiveness of using improved residual networks for palmprint recognition,which is of great significance for using deep learning technology for palmprint recognition.