文献类型：期刊文献

中文题名：改进半监督GAN及在糖网病分级上的应用

英文题名：Improved semi-supervised GAN and its application in classification of diabetic retinopathy

作者：岳丹阳[1];罗健旭[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2022

卷号：43

期号：8

起止页码：2204

中文期刊名：计算机工程与设计

外文期刊名：Computer Engineering and Design

收录：CSTPCD;;北大核心:【北大核心2020】；

基金：上海市科技创新行动计划基金项目(19511121203)。

语种：中文

中文关键词：半监督分类;半监督生成对抗网络;三重生成对抗网络;残差网络;糖网病

外文关键词：semi-supervised classification;semi-supervised generative adversarial network;Triple-GAN;residual network;diabetic retinopathy

摘要：针对糖网病标注数据量少的问题,提出一种改进的半监督生成对抗网络方法,可以利用大量无标签数据和生成数据提升分类精度。基于Triple-GAN的半监督学习方法,提出多尺度残差网络,使用多层特征提高分类精度;设计一种压缩激活双注意力生成对抗网络,生成具有大范围关联性的图像,提高半监督模型的数据分布拟合能力。实验验证,提出模型的分类精度优于常见一些半监督模型,在糖尿病眼底图像上取得了较高的分类精度。
As to the problem of the small amount of labeled data for diabetic fundus lesions,an improved semi-supervised generative adversarial network,which used large amounts of unlabeled data and generated data to enhance the accuracy of classification,was proposed.Based on the semi-supervised learning method of Triple-GAN,a multi-scale residual network was proposed,which used multi-layer features to improve classification accuracy.Meanwhile,a squeeze-and-excitation dual-attention generative adversarial network that produced large-range relevant images was designed to improve the data distribution fitting ability of the semi-supervised model.Experimental verification show that,the classification accuracy of the proposed model is better than some common semi-supervised models,and higher classification accuracy is achieved on the diabetic fundus images.