Impervious surface is an important factor in monitoring urban development and environmental analysis. However, spectral differences and structural differences exist on impervious surfaces, which leads to accurate extraction of impervious surfaces is a difficult task. Therefore, this paper proposes a superpixel sparse representation based on morphological profiles and raw data to extract the impervious surface on hyperspectral imagery. Specifically, the segmentation map of the image is first generated by the mean shift segmentation. Then, the morphological attribute profiles of the hyperspectral image are extracted and stacked with raw data. Finally, the segmentation map is masked onto the stacked image and each resulting superpixel is classified via sparse representation. Experiments show that the method has good performance in impervious surface extraction and has advantages in the comparison method. This shows the effectiveness of the proposed method.
展开▼
