基于轻量级U-Net神经网络提取OCT内指纹和外指纹  

OCT internal and external fingerprints generation based on a lightweight U-Net neural network

作　　者：王海霞[1] 杨熙丞 梁荣华[1] 陈朋[1] Wang Haixia;Yang Xicheng;Liang Ronghua;Chen Peng(College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310000;College of Information Engineering,Zhejiang University of Technology,Hangzhou 310000)

机构地区：[1]浙江工业大学计算机科学与技术学院,杭州310000 [2]浙江工业大学信息工程学院,杭州310000

出　　处：《高技术通讯》2019年第11期1063-1072,共10页Chinese High Technology Letters

基　　金：国家重大科研仪器研制项目(61527808);国家自然科学基金(61602414);浙江省自然科学基金(LY19F050011);浙江省基础公益研究计划(LGG19F020011)资助项目

摘　　要：光学相干断层扫描(OCT)技术是一种非侵入式的成像技术,可以用来采集高分辨率的手指3维数据,提取角质层和乳头层轮廓并生成内指纹和外指纹。针对目前已有算法在提取角质层和乳头层轮廓时易受到汗腺和皮下组织的影响导致轮廓提取结果有偏差这一问题,本文利用自制OCT实验平台获取高分辨率手指3维数据,提出了一种基于深度可分离卷积的轻量级U-Net神经网络算法来准确提取角质层和乳头层轮廓,通过拼接轮廓的相对深度信息生成内指纹和外指纹。实验结果表明,本文提出的算法能够精确地提取内外指纹,同时在生成指纹效果和普通U-Net神经网络算法相似的前提下大幅减少了模型参数数量。Optical coherence tomography(OCT)is a non-destructive imaging technique which has been used to acquire high-resolution 3D data of fingertips from which contours of stratum corneum and papillary are extracted and subsequently internal and external fingerprints are generated.Since most current contour extraction methods are susceptible to sweat glands and subcutaneous tissues,the estimated stratum corneum and papillary contours may be deviated from its real location.In this paper,a self-built OCT system is used to obtain high-resolution 3D fingertip data and a lightweight U-Net neural network based on depth separable convolution is proposed to accurately extract the stratum corneum and papillary contours.The depth information of these contours are then used to generate internal and external fingerprints.The experimental results show that the proposed method can accurately extract internal and external fingerprints.Compared to the original U-Net neural network,the proposed lightweight U-Net greatly reduces the number of model parameters while obtaining similar fingerprint generation results.

关 键 词：光学相干断层扫描(OCT) 内指纹 外指纹 深度可分离卷积 U-Net 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]