Research On Fingerprint Classification Application Based On The Lightweight Neural Network

Fingerprint classification is not only a very important technology for target fingerprints searching in large scale fingerprint database,but also a crucial part in the process of citizen verification and criminal searching in some countries.It's difficult for traditional fingerprint classification method to get a perfect performance when it has low adaptability to fingerprints with different quality and specifications,especially in manual calculation.Recently,deep learning is widely used in various image processing because of its powerful expression ability,to this regard,the network is trained by large scale database with tags can autonomously extract abstract fingerprint features,which explores a new direction for fingerprint classification that makes it adaptive and less manual intervention.Thus,the lightweight neural network is construct in this paper for automatic features extraction and fingerprint images classification.Firstly,various feature fusion methods are proposed to enhance the representation ability of neural networks;Secondly,transfer learning is introduced to reduce the requirements of the target data scale and improve the classification performance on small sample fingerprint dataset,The main contents are as follows:(1)A fingerprint classification model based on lightweight Finger-SqueezeNet network is purposed.Since the fingerprint classification dataset is small in scale,it will occur over-fitting if the deep convolutional neural network is trained directly.Therefore,the paper adopts a lightweight network instead of a deep convolutional neural network to reduce the parameters and speed up the model to converge.It builds a Finger-SqueezeNet model based on the SqueezeNet network,which greatly reduces the number of parameters.(2)A lightweight fingerprint classification model based on transfer learning is proposed.In order to improve the classification performance of small sample fingerprint images,this paper combines the idea of transfer learning on the basis of lightweight network.Firstly,the fingerprint orientation filed database is used to pre-train the lightweight network.Then,the fingerprint data set is used to optimize the parameters.Finally,the optimized network is verified by fingerprint classification experiment.The experimental results show that the lightweight network model of migration can effectively improve the fingerprint classification performance.(3)A fingerprint classification algorithm based on lightweight multi-feature fusion is put forward.Firstly,the fingerprint image is input into the lightweight network to learn and extract the depth feature.At the same time,the refined image of the fingerprint is obtained by using the look-up table method,and the improved distributed summation gradient method is used to get the Region of Interest(ROI).Then,the fingerprint ROI is merged with the deep features,so that the deep network can also learn the shallow ridge trend information,thereby,the network sensitivity to the pattern is enhanced.The experimental results show that the fingerprint ROI is merged in the deep features can fully learn the fingerprint pattern information to improve the fingerprint classification results.