基于类别扩展的广义零样本图像分类方法  

Category Expansion Based Generalized Zero-Shot Image Classification

作　　者：张杰 廖盛斌 张浩峰[1] 陈得宝 ZHANG Jie;LIAO Sheng-bin;ZHANG Hao-feng;CHEN De-bao(School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing,Jiangsu 210094,China;National Digital Learning Engineering Technology Research Center,Central China Normal University,Wuhan,Hubei 430079,China;School of Computer Science and Technology,Huaibei Normal University,Huaibei,Anhui 235000,China)

机构地区：[1]南京理工大学计算机科学与工程学院,江苏南京210094 [2]华中师范大学国家数字化学习工程技术研究中心,湖北武汉430079 [3]淮北师范大学计算机科学与技术学院,安徽淮北235000

出　　处：《电子学报》2023年第4期1068-1080,共13页Acta Electronica Sinica

基　　金：国家自然科学基金(No.61872187,No.62077023,No.62072246);江苏省自然科学基金(No.BK20201306)。

摘　　要：在传统的零样本图像分类方法中,语义属性通常被用作辅助信息来描述各类别的视觉特征.然而,单一的语义属性并不能对类内多样性的视觉特征进行全面的描述.为提高语义属性对类别内部多样性的表示能力,同时也为了帮助模型提高对各类别的描述能力,本文通过属性自编码器的方式在视觉以及语义空间上对类别进行扩展.此外,为了缓解传统生成性方法因无法直接计算生成空间到真实空间的变换而带来的模型次优解问题,本文采用了生成流网络作为基础网络,通过可逆变换直接计算两个空间之间的变换来开展对零样本学习任务的研究.本文使用解码器网络将逆生成流网络为测试样本生成的原型特征解耦成视觉原型及语义原型信息,然后根据这两个原型信息实现将测试样本预分类到可见类集或不可见类集中,最终在这两个子分类空间中根据样本的特点分别进行监督分类和零样本分类任务以提高模型的整体性能表现.本文在五个数据集上通过大量的实验验证了本文所提方法的有效性.In traditional zero-shot image classification methods,semantic attributes are usually used as auxiliary in-formation to describe the visual features of each class.However,a single semantic attribute cannot fully describe the diverse visual features within a single class.To improve the ability of semantic attributes to express the diversity within the class,and to help the model improve the description ability for each category,we use the semantic auto-encoder to expand the cat-egories in visual and semantic space.In addition,to alleviate the suboptimal solution problem of the model caused by the in-ability to directly calculate the transformation from the generation space to the real space by the traditional generative meth-ods,we employ the generative flow as the basic network in this paper to directly calculate the transformation between the two spaces.Furthermore,we exploit the decoder network to decouple the prototype features generated by the inverse genera-tive flow network for the test samples into visual prototypes and semantic prototypes,and then realize the pre-classification of the test samples into seen or unseen classes.Finally,in the two sub-classification domains,supervised classification and zero-shot classification are performed separately to improve the overall performance.Extensive experiments are conducted on five popular datasets to verify the effectiveness of the proposed method.

关 键 词：生成流 预分类 广义零样本学习 类扩展 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP37[自动化与计算机技术—计算机科学与技术]