An Efficient PAN-sharpening of Multispectral Images using Multi-scale Residual CNN with Sparse Representation

In order to create "Multi-Spectral (MS)" images with high spatial resolution, the PAN-sharpening method is used, which combines spectral data from the MS image with spatial data from the "panchromatic image". Spectrum distortion is the greatest issue in traditional methods while obtaining a sharpened MS image with good spatial resolution. To convert color space to Intensity-Hue-Saturation (IHS) and offer equalization of a Panchromatic (PAN) component with an MS image, IHS methods are used, and this method minimizes distortion issues. Majority of IHS techniques still result in spectrum distortion, which is very difficult to handle. To crack this crisis, a narrative "PAN sharpening of MS Images" is implemented using deep learning to get super-resolution images. Initially, the PAN as well as MS images are taken from various online sources. The sparse representation from the combined images is obtained to get more matching patterns from the images. From this sparse representation, more stimulated output is retained. This stimulated output is combined with the PAN images and the resultant outcome is subjected to the "Multi-scale Residual Convolutional Neural Network (MSRCNN)" for getting the super-resolution images.