山西大学学报（自然科学版）2025，Vol.48Issue(6)：1161-1170,10.DOI:10.13451/j.sxu.ns.2024055
近邻一致性策略下的图像深度聚类算法研究
Research on Image Deep Clustering Algorithm Based on Near Neighbor Consistency Strategy
摘要
Abstract
Image clustering is an important task in the field of computer vision.Although many methods have been proposed to solve the image clustering task,current deep clustering methods based on representation mainly rely on the discriminative representation ca-pability of samples.However,in addition to discriminative representation,the consistency of sample distribution probabilities with their nearest neighbor samples should also be considered so that the learned sample representation space possesses discriminative,sta-ble,and consistent properties.Based on this,this paper conducts research on image deep clustering algorithms using the strategy of nearest neighbor consistency.The method consists of two stages:the representation learning stage,where a convolutional autoencoder is trained to construct the initial feature space,and the clustering stage,where nearest neighbor consistency is used as a constraint and sample stability is incorporated as an enhancement to fine-tune the network based on the first stage,obtaining the final clustering distri-bution.This method primarily considers the consistency of probability assignment between each sample and its nearest neighbors dur-ing clustering,fully exploring the similarity relationships among samples of the same class to achieve a compact sample distribution.Experimental results demonstrate that the proposed method outperforms typical clustering algorithms on five image datasets.关键词
深度聚类/图像/自编码器Key words
deep clustering/image/autoencoder分类
计算机与自动化引用本文复制引用
钱宇华,程占文,李飞江,王婕婷..近邻一致性策略下的图像深度聚类算法研究[J].山西大学学报（自然科学版）,2025,48(6):1161-1170,10.基金项目
国家自然科学基金(62106132 （）
62306170) （）
山西省科技重大专项(202201020101006) （202201020101006）
山西省基础研究计划(20210302124271 （）
202103021223026) （）
山西省科技创新人才团队专项资助(202304051001001) （202304051001001）
