基于关系有向图正则化的属性三因子分解模型  

Attribute Tri-factorization Model with Regularization of Relation Digraph

作　　者：张嘉睿 李瑞林[3] 孔毅 余南南 ZHANG Jiarui;LI Ruilin;KONG Yi;YU Nannan(School of Electrical Engineering and Automation,Jiangsu Normal University,Xuzhou 221116,China;School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116,China;State Key Laboratory of Intelligent Construction and Healthy Operation&Maintenance of Deep Underground Engineering,China University of Mining and Technology,Xuzhou 221116,China)

机构地区：[1]江苏师范大学电气工程及自动化学院,江苏徐州221116 [2]中国矿业大学信息与控制工程学院,江苏徐州221116 [3]中国矿业大学深地工程智能建造与健康运维全国重点实验室,江苏徐州221116

出　　处：《郑州大学学报（理学版）》2025年第1期67-73,共7页Journal of Zhengzhou University:Natural Science Edition

基　　金：国家自然科学基金项目(42372329);江苏省高等学校基础科学(自然科学)研究项目(21KJB520005);江苏省自然科学基金项目(BK20200632);徐州市基础研究计划(KC23019);江苏师范大学科研项目(21XSRS001)。

摘　　要：针对零样本图像分类中属性和特征之间映射不全面以及属性空间结构挖掘不充分问题,提出了基于关系有向图正则化的属性三因子分解模型。首先,利用属性矩阵三因子分解实现了属性空间和特征空间的映射;其次,通过权值矩阵构建了属性关系有向图;最后,在属性空间或特征空间计算测试样本和各测试类别的相似性,进而实现图像分类。在aPY和SUN数据集上的实验结果表明,所提模型有效地提高了零样本图像分类精度。Aiming at the problems of incomplete mapping between attributes and features,as well as the insufficient mining of the attribute space structure in zero-shot image classification,an attribute tri-factorization model with regularization of relation digraph was proposed.Firstly,the mapping between attribute space and feature space was achieved by matrix tri-factorization of attributes.Secondly,the attribute relation digraph was constructed using the weight matrix.Finally,the similarity between the testing sample and each testing class was calculated in either the attribute space or the feature space to finish image classification.Experimental results on aPY and SUN datasets showed that the proposed model was capable of efficiently improving the accuracy of zero-shot image classification.

关 键 词：零样本图像 属性三因子分解 关系有向图 正则化 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]