基于低秩重构的迁移子空间学习物体图像识别方法OA

Transfer subspace learning based on low-rank reconstruction is an effective cross-domain learning method,but most of existing methods only consider the classification interval between categories and lose some discriminative information,and are prone to overfitting.To address these issues and enhance the classification performance of the model,this paper proposes a low-rank reconstruction based transfer subspace learning(LRR-TSL)method.On the basis of transferring subspaces,low-rank constraints are used to ensure alignment between the source and target domains within the subspace,and sparse constraints are used to ensure reconstruction from the source domain to the target domain.A relaxed label matrix is introduced and combined with the ε-draggings technique to improve the intra-class compactness and inter-class separation of the model,while avoiding overfitting of the relaxed labels.The linear entropy based on the reconstruction matrix and the label matrix is constructed to further enhance the discriminative ability of the model.Cross-domain image recognition experiments on Office and Caltech-256 datasets demonstrate the effectiveness of the proposed method in this paper.