Sparse fusion based on SAM elective sample dictionary establishment

In the paper, it is proposed that multi-spectral image pure pixel is utilized for completing SAM classification. The classified samples are utilized for electively constructing sparse dictionary, thereby improving the representativeness of the dictionary. Eight surface feature types are set in Landsat8 image. PPI index is used for calculating pure pixel index of each pixel. Pure pixel of each surface feature is further extracted through N-D visualizer, which is used for SAM calculation. Eight kinds of surface feature samples are selected from SAM image for online dictionary learning. Multi-spectral image sparse dictionary is generated. Multi-spectral image sparse coefficient is calculated through dictionary and OMP. Meanwhile, online dictionary and OMP are utilized for obtaining panchromatic image sparse coefficient. Fusion sparse coefficient is generated by maximum values both sparse coefficients. Multi-spectral image sparse dictionary is combined for reconstructing and generating fusion image. Eight quantitative fusion evaluation indicators are adopted for comparing algorithm fusion and weighted fusion in the paper. Fusion method proposed in the paper contains more information, fusion image texture detail information is improved, and better image multi-spectral information is kept.