基于成对约束的稀疏嵌入遥感影像降维

iming at the huge amount of remote sensing data with little labeled samples, a novel dimensionality reduction approach is introduced, which is named sparsity preserving with pair-wise constraints for dimensionality reduction.It considers labeling samples by the use of positive and negative constraints. Pair-wise constraints take an advantage of having a better classification performance and being easy to get them. Sparse preserving projection belongs to unsupervised dimensionality reduction, so the samples' supervised information is not fully utilized. Therefore, the algorithm introduces positive constraints and negative constraints to guide sparse reconstruction. The both will be effectively fused, namely using positive constraints and negative constraints as the supervision of information to guide sparse reconstructed for improving the classification effect. The results on remote sensing datasets verify the feasibility and validity of the proposed methods.