Downscaling Multispectral Imagery Based on the HNN Using Forward Model

Super-resolution techniques can be used to increase the spatial resolution of the imagery. Although widely applied in image processing, the current super-resolution approaches were hardly applicable for super-resolution of remotely sensed multispectral (MS) imagery because of the lack of a sequence of images in a scene at the same time. In this paper, three methods used for increasing the spatial resolution of a MS image are introduced. The paper is divided into two parts. In part I, a new model for predicting the MS image at the sub-pixel spatial resolution is presented. The new model is based on the super-resolution mapping technique, soft-classification and a forward model using local end-member spectra. Soft-classification was based on unsupervised classification to reduce the land cover proportion prediction error. The statistics used for image assessment were Root mean squares error (RMSE). Both the visual and statistical evaluation showed that the new methods can generate adequate MS images with a finer spatial resolution than that of the original imagery. Visually, the super-resolved image was apparently sharper than the original coarse spatial resolution image. Similarly, the RMSEs of the super-resolved image compared to the reference were smaller than those of the original degraded image due to the super-resolution of mixed pixels. I