A simple and effective spectral-spatial method for mapping large-scale coastal wetlands using China ZY1-02D satellite hyperspectral images

Abstract This paper proposes a simple and effective spatial-spectral (SESS) method for mapping large-scale coastal wetlands using China ZY1-02D satellite hyperspectral data. First, the improved sparse subspace clustering algorithm is implemented to select presentative bands, and then the multi-scale low-rank decomposition algorithm is employed to extract multi-scale spatial features. After that, the extracted spatial and selected spectral features are stacked and fused using presented Landmark-neighborhood preserving embedding algorithm. Finally, the low-dimensional fused features are classified with two popular classifiers (i.e., random forest and support vector machine) to map coastal wetlands. Experiments results show that the SESS obtains the highest classification accuracy of 96.92% and 94.84% on Yellow River Delta and Yancheng coastal wetlands of China, respectively, exhibiting good capability in accurately identifying complicated ground objects. This paper is the first attempt to investigate the potentials of China ZY1-02D satellite hyperspectral data, and the SESS can also be applied into other hyperspectral data.