超像元尺度下的多光谱图像超分辨率洪水淹没制图

超分辨率制图SRM(Super-resolution Mapping)技术可以有效地处理遥感图像中的混合像元,获得准确的地物类别分布信息.目前,SRM技术已经成功地应用于多光谱图像洪水淹没定位中,称为超分辨率洪水淹没制图SRFIM(Super-resolution Flood Inundation Mapping).然...

Super-resolution flood inundation mapping for multispectral image based on super-pixel scale

Super Resolution Mapping (SRM) technology can effectively handle mixed pixels in remote sensing image and obtain the accurate distribution information of land-cover class. SRM technology is currently successfully applied to flood inundation mapping for multispectral image, which is called Super Resolution Flood Inundation Mapping (SRFIM). However, the existing SRFIM methods are often based on pixel-scale spatial correlation. This method considers the spatial relationship between pixels in the set rectangular window, but the shape of the inundation area or the non-inundation area is irregular in reality. Thus, the pixel-scale spatial correlation is insufficiently accurate, which affects the final accuracy of flood inundation mapping. Super-resolution flood inundation mapping of multispectral image based on super-pixel scale spatial correlation (SSSC-SRFIM) is proposed to solve the abovementioned problem.In SSSC-SRFIM, the original coarse multispectral image is first improved by bicubic interpolation to obtain the improved image, and the fractional image with the proportion value of each subpixel belonging to inundation area is obtained by unmixing the improved image. The first principal component of the improved image is then extracted by principal component analysis, and the image segmentation based on multi-resolution is used to segment the first principal component to obtain the super-pixels with irregular shape. Next, the fractional image and super-pixels are integrated, and the random walk algorithm is introduced to calculate the super-pixel-scale spatial correlation. Finally, according to the super-pixel-scale spatial correlation, the label of the inundation area or the non-inundation area is assigned to each sub-pixel by the class allocation method based on the unit of object. Thus, the final result of flood inundation mapping is produced.Two Landsat 8 OLI multispectral images are used to evaluate the method. The proposed SSSC-SRFIM method has better performance than the traditional methods.In the proposed SSSC-SRFIM, the super-pixel-scale spatial correlation is more accurate than pixel-scale spatial correlation because the irregular distribution shape of the actual inundation and non-inundation areas is considered. Therefore, better flood inundation mapping result can be obtained by the proposed SSSC-SRFIM.