Feature Selection for Cross-Scene Hyperspectral Image Classification Using Cross-Domain I-ReliefF

In the classification of hyperspectral images (HSIs), too many spectral bands (features) cause feature redundancy, resulting in a reduction accuracy. order to solve this problem, it is good method use selection search for subset which useful classification. Iterative ReliefF (I-ReliefF) traditional single-scene-based algorithm, and has convergence, efficiency, can handle problems well most scenes. Most methods perform poorly some scenes (domains) lack labeled samples. As number HSIs increases, cross-scene algorithms utilize two deal with high dimension low sample size problem are more desired. The shift common selection. It leads difference distribution between source target even though these highly similar. To above problems, we extend I-ReliefF algorithm: cross-domain (CDIRF). CDIRF includes rule update weights, considers separability different land-cover classes consistency features So effectively information scene improve performance scene. experiments conducted on three datasets verification, experimental results demonstrate superiority feasibility proposed algorithm.