一种结合类激活映射的半监督图像分类方法  

Semi-supervised Image Classification Method Combined with Class Activation Mapping

作　　者：王宪保[1] 肖本督 姚明海[1] WANG Xian-bao;XIAO Ben-du;YAO Ming-hai(College of Information Engineering,Zhejiang University of Technology,Hangzhou 310023,China)

机构地区：[1]浙江工业大学信息工程学院,杭州310023

出　　处：《小型微型计算机系统》2022年第6期1204-1209,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61871350)资助;浙江省科技计划项目(2019C011123)资助;浙江省基础公益研究计划项目(LGG19F030011)资助.

摘　　要：半监督学习要求无标记数据集远大于标记数据集,然而无标记数据集中包含的复杂无关信息又会对模型训练造成负面影响.针对此问题,本文提出了一种基于增强的均值教师模型的半监督图像分类方法.首先根据类激活映射的工作机制,构建一个具有类激活映射功能的网络;然后将无标记数据集输入结合类激活映射的目标初定位网络,得到目标初定位图;最后将标记图像和目标初定位图像组成训练数据集,训练得到半监督图像分类器.本文设置了标记数据占比、骨干网络、数据集的对比实验,结果表明,本文算法在Top1和Top5上的表现优于现有算法,说明了本文算法的可行性和有效性.Semi-supervised learning requires that the unlabeled dataset is much larger than the labeled dataset.However,the complex and irrelevant information contained in the unlabeled dataset will negatively affect model training.Aiming at this problem,this paper proposes a semi-supervised image classification method based on the enhanced mean teacher model.First,according to the working mechanism of class activation mapping,construct a network with the function of class activation mapping;Then input the unlabeled dataset into the target preliminary localization network combined with class activation mapping to obtain the preliminary localization map of the target;Finally,the labeled image and the target preliminary localization image are formed into a training dataset,and a semi-supervised image classifier is trained.This paper sets up a series of comparative experiments on the proportion of labeled data,backbone network,and datasets.The results show that the performance of this algorithm is better than existing algorithms on Top1 and Top5,which shows the feasibility and effectiveness of this algorithm.

关 键 词：半监督学习 图像分类 卷积神经网络 类激活映射 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]