Considering Point Spread Function Effect for Spectral Remote Sensing Image Super-Resolution Mapping

‍ ‍Because of hardware limitations and the diversity of land-cover classes， the spatial resolution of collected spectral remote sensing images is sometimes coarse， producing many mixed pixels， which seriously affects the accuracy when mapping the spatial distribution of land-cover classes. Super-resolution mapping technology can effectively process the mixed pixels in a spectral remote sensing image and obtain accurate information on the distribution of land-cover classes. In the context of remote sensing big data， multiple sub-pixel shifted images collected by the same satellite in the same area can be used as auxiliary data to improve the super-resolution mapping results. However， the current methods used for super-resolution mapping in multiple sub-pixel shifted-image interpolation are rarely effective in considering the effect of the point spread function， which reduces the accuracy of the mapping results. To address this issue， this paper proposes a method that considers the point-spread function effect for spectral remote sensing image super-resolution mapping. In the proposed method， multiple sub-pixel shifted images are first unmixed to produce coarse fractional images. Then， while considering the point-spread function effect， improved coarse fractional images are obtained by applying the area-to-point kriging and then the ideal square-wave filter to the coarse fractional images. Next， the improved coarse fractional images are upsampled by interpolation to obtain upsampled fractional images， which are then integrated to obtain fine fractional images. Finally， based on the class proportion information from the fine fractional images， the class allocation method is utilized to assign class labels to sub-pixels to obtain the ideal mapping results. Experimental results showed that the proposed method exhibited the optimal performance. For example， the overall accuracy and Kappa coefficient of the experimental results for the Washington， D.C.， USA Dataset reached 77.63% and 0.7263， respectively.