Research On An Optimization Of Deep Learning Model For Fingerprint Direction Recognition

With the development of technology and popularization of various electronic information equipment,people have higher and higher requirements for their own information security.The cumbersome symbolic password authentication method can no longer adapt to the increasingly accelerated social rhythm,and there is an urgent need for more security 3.More convenient means of identity verification.Fingerprint recognition has always been valued as an ancient biometric identity verification method.People have developed automatic fingerprint recognition technology based on fingerprint recognition theory and combined with modern information technology and tools.In the traditional fingerprint recognition algorithm,we think that it is unreasonable to only use the detailed feature points of the fingerprint as the basis for fingerprint recognition,and it is unreasonable to ignore the fingerprint information.The fingerprint direction field not only contains the information of the fingerprint fingerprint direction but also implies In addition to other rich feature information,the realization of fingerprint recognition in combination with the fingerprint direction field is a reflection of the full use of fingerprint features,and it will also further improve the accuracy of the final fingerprint recognition.The theory of deep learning has developed rapidly in recent years,and the more the application More and more widely,it solves many difficult problems in classic information processing.Under this background,some researchers have proposed a fingerprint direction field recognition method based on deep learning.They use the idea of full connection and dense connection to build a network model.Good results have been achieved,however,the generated model is too large and the model accuracy needs to be improved.In order to more fully extract the characteristics of the fingerprint direction field data,we use the two-dimensional data form of the fingerprint direction field as the input of the network,and introduce a convolution structure to reduce the use of model parameters to a certain extent.In addition,the fingerprint direction field used in the analysis The data is one-shot-data,and the use of general classification networks is not enough to achieve such a huge number of categories,and the amount of data in each category is very small.For this reason,we use metric learning methods to convert classification problems into clustering problems.Construct a twin network model.Extract networks for different features successively.Experimental comparison.Finally,a network with higher accuracy and smaller parameters is used.By analyzing the output of each hidden layer channel and the last feature channel,we cut some channels.To further simplify the model.Experiments show that our network model can not only further improve the accuracy of fingerprint direction field recognition,but also reduce the size of the model,which achieves our expected effect.Not only shows that the idea of using fingerprint direction field for fingerprint recognition is feasible,And can be further improved.