ISSN 1000-1239 CN 11-1777/TP

计算机研究与发展 ›› 2018, Vol. 55 ›› Issue (8): 1667-1673.doi: 10.7544/issn1000-1239.2018.20180215

所属专题: 2018数据挖掘前沿进展专题

• 人工智能 • 上一篇    下一篇

基于卷积神经网络的左右眼识别

钟志权1,2,袁进3,唐晓颖4   

  1. 1(广东顺德中山大学卡内基梅隆大学国际联合研究院 广东顺德 528300);2(中山大学电子与信息工程学院 广州 510006);3(眼科学国家重点实验室(中山大学) 广州 510060);4(南方科技大学电子与电气工程系 广东深圳 518055) (zhongzhq9@mail2.sysu.edu.cn)
  • 出版日期: 2018-08-01
  • 基金资助: 
    国家重点研发计划项目(2017YFC0112400);国家自然科学基金项目(81501546) This work was supported by the National Key Research and Development Program of China (2017YFC0112400) and the National Natural Science Foundation of China (81501546).

Left-vs-Right Eye Discrimination Based on Convolutional Neural Network

Zhong Zhiquan1,2, Yuan Jin3,Tang Xiaoying4   

  1. 1(SYSU-CMU Shunde International Joint Research Institute, Shunde, Guangdong 528300);2(School of Electronics and Information Technology, Sun Yat-sen University, Guangzhou 510006);3(State Key Laboratory of Ophthalmology (Sun Yat-sen University), Guangzhou 510060);4(Department of Electrical and Electronic Engineering, Southern University of Science and Technology, Shenzhen, Guangdong 518055)
  • Online: 2018-08-01

摘要: 利用一个深度卷积神经网络提出并验证了一种能自动判别左右眼的新方法.根据左眼和右眼图像的特征差异性,所设计的网络能够自动估计网络的所有参数.在Alexnet网络的基础之上进行修改,设计的卷积神经网络由4个卷积池化层和2个全连接层组成,其次是作为最后一层的分类器.基于针对42541张眼底图像的实验结果,所提出的网络的训练精度约为100%,测试精度高达99%.此外,所提出的网络具有高度的鲁棒性,它可以成功地识别大量具有多变性的眼底图像.据所知,这是第1个基于深度学习用于左右眼识别的高精准度网络.

关键词: 左右眼识别, 眼底图像, 深度学习, 卷积神经网络, Alexnet

Abstract: In this paper, a new method to automatically discriminate the left and right eyes is proposed and validated utilizing a deep convolutional neural network. All parameters of the designed network are automatically estimated based on the characteristic differences between the left and right eye images. On the basis of the Alexnet network, the convolutional neural network designed in this paper consists of four convolutional layers and two fully connected layers, followed by a classifier serving as its last layer. According to our experimental results on a total of 42541 fundus images, the training accuracy of our network is about 100%, and the testing accuracy is as high as 99%. In addition, the proposed network is highly robust given that it successfully works for a large amount of fundus images with great variability. As far as we know, this is the first deep learning based network for left-vs-right eye discrimination that exhibits very high accuracy and precision.

Key words: left-vs-right eye discrimination, fundus images, deep learning, convolutional neural network (CNN), Alexnet

中图分类号: