基于ResNet的彩色眼底图片分类算法研究  

RESEARCH ON CLASSIFICATION ALGORITHM OF COLORFUNDUS IMAGES BASED ON RESNET

作　　者：陈亚浩 张东[2] Chen Yahao;Zhang Dong(School of Information Engineering,Henan Supercomputing Center of Zhengzhou University,Zhengzhou 450000,Henan,China;Inspur Electronic Industry Co.,Ltd.,Jinan 250101,Shandong,China)

机构地区：[1]郑州大学河南省超级计算中心信息工程学院,河南郑州450000 [2]浪潮电子信息产业股份有限公司,山东济南250101

出　　处：《计算机应用与软件》2023年第8期250-254,320,共6页Computer Applications and Software

摘　　要：为了提高现有算法对眼底彩照的识别准确度,提出一种基于深度残差网络(ResNet)的眼底图像分类方法,对获取到的眼底图像进行基于DSP-Fs流程的数据预处理操作,使用Laplacian滤波处理以突出异常眼底图像的特征,有效地提高神经网络学习的质量;用残差网络代替传统的卷积神经网络,提取更深层次的特征,同时修改网络结构以提升模型效率达到病变分类的目的。在ODIR数据集上对不同的预处理方式和网络结构进行了对比实验,结果表明,该算法能够有效地提升眼底彩色图像分类的准确度。In order to improve the recognition accuracy of fundus color photos by existing algorithms,this paper proposes a fundus image classification method based on deep residual network(ResNet).The acquired fundus images were preprocessed based on the DSP-Fs process.Laplacian filtering was used to highlight the characteristics of abnormal fundus images and effectively improve the quality of neural network learning.The residual network was used to replace the traditional convolutional neural network to extract deeper features,and the structure of the network was modified for classification to improve model efficiency and achieve the purpose of lesion classification.Comparative experiments on different preprocessing methods and network structures on the ODIR data set show that the algorithm can effectively improve the accuracy of fundus color image classification.

关 键 词：眼底图像 图像分类 残差网络 Laplacian滤波 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]