使用模糊标签驱动标签松弛的多视角分类算法OA

With the continuous advancement of data acquisition technologies,multi-view data has been widely applied in fields such as medical imaging,action recognition,and multimodal analysis.However,semantic discrepancies across different views and subjectivity in the label annotation process often lead to label noise and reduced classification robustness.To address these issues,this paper proposed a multi-view classification algorithm that integrates fuzzy-label-driven label relaxation with consistency regularization.Specifically,fuzzy clustering was employed to construct soft label representations for each sample,aiming to capture the uncertainty and semantic ambiguity in the labels.During the learning of soft labels,the method introduced view-specific weights and joint constraints with ground-truth labels to guide the model in establishing a flexible supervision mechanism between hard and fuzzy labels,thereby achieving a smooth transition at the label level.Finally,through iterative optimization of both soft labels and view features,the model learned a discriminative feature projection matrix.Experimental results on four real-world multi-view datasets demonstrated that the proposed method outperforms conventional multi-view and traditional classification approaches in terms of robustness to label noise and effectiveness in multi-view information integration.