文献类型：期刊文献

中文题名：基于生成对抗网络的图像识别改进方法

英文题名：Improved method for image recognition based on generative adversarial network

作者：李凯[1];彭亦功[1]

机构：[1]华东理工大学信息科学与工程学院,上海200237

年份：2019

卷号：40

期号：2

起止页码：492

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2017】；

语种：中文

中文关键词：图像识别;不清晰图像;深度学习;生成对抗网络;生成模型

外文关键词：image recognition;unclear images;deep learning;generative adversarial network;generative model

摘要：针对低分辨率、不清晰图像识别精度较低问题,提出基于生成对抗网络(GAN)的图像识别改进方法。通过GAN中的生成模型与判别模型之间的极大极小博弈,使生成模型获得修复不清晰图像的能力,由此生成模型与一般分类网络相结合生成新网络,可从不清晰图片中提取准确的特征,提高对不清晰图像的识别精度。实验结果表明,改进方法对不清晰图像的识别精度有显著提升,对提高图像识别质量具有重要的价值。
Aiming at the problem of low resolution and unclear image recognition accuracy,an image recognition approach based on generative adversarial network(GAN)was presented,in which the generative model and the discriminative model played a minimax two-player game and the generative model got the ability of repairing unclear images.The new network that combined the generative model and the classification network can extract accurate features from the unclear images to improve the recognition accuracy of unclear images.Experimental results show that the improved approach is more accurate in recognizing unclear images,which is of excellent value to improve the quality of image recognition.