Pansharpening of Multispectral Images using Dual Attention based Two Stream Network

Abstract

With the advent of technology there are large number of earths observing satellites having sensors, delivering different type of images that can provide a varying useful information about different surfaces of earth and those areas that are not reachable by human. Most popular of these images are panchromatic (PAN) image having the geometric details and multispectral (MS) image that contains reflectance data that represents wide variety of spectral band and wavelengths. Availability of these image encouraged the development and utilization remote sensing application. Regardless of being rich in spectral details, MS images lack in spatial information due to the sensors limitations. Pansharpening approach is devised to enhance the spatial details of MS image using the spatial features of PAN image. This method simply takes the spatial information of PAN and fuse it with MS im age and produce a pansharped MS image high in both spectral and spatial information. In this thesis, we explored literature review and following the new trends we proposed a new dual attention based two stream pansharpen ing approach that extracts the details from both MS and PAN images and fuse them together to produce a pansharped MS image. For analysis we performed different experiments using Pleaides dataset and compared our model with a recently proposed approach with the help of visual results and different quality metrics.