基于对比学习的深度嵌入图像聚类算法OA

Contrastive clustering is an unsupervised learning method that combines contrastive learning with clustering.The method utilizes the similarities and differences between samples to extract useful feature representations,which improves the clustering effect and the discriminative ability of the model in unsupervised learning tasks.However,existing comparative clus-tering algorithms fail to fully retain the original data information,resulting in the loss of information in the embedding space,which limits the abillity of expression and effectiveness of the feature representation.In addition,the existing algorithms focus more on the construction of positive and negative sample pairs,and lack the optimized design for the clustering task,thus mak-ing it difficult to learn feature representations that are truly suitable for the clustering task.To address these issues,this paper proposes a deep embedded image clustering algorithm based on contrast learning.The algorithm digs deeply into the multilevel re-lationship between samples through instance-level and cluster-level contrast learning,and effectively improves the stability of the model in the clustering task.Meanwhile,by jointly optimizing the reconstruction loss of the self-encoder and the constraints of the deep embedding representation,the model is able to retain more original sample features and enhance the discriminative ability of the embedding representation to ensure that the key information in the embedding space is effectively retained,thus im-proving the clustering performance.Experimental results show that the method in this paper demonstrates excellent performance in tests on five public datasets,which fully validates the effectiveness of the algorithm.