Designing algorithms that spatially enhance hyperspectral imagery is an active area of research due to their many wide-ranging applications such as target detection, precision agriculture, forest management, security/defense applications, among others. This body of work focuses on the effects of spatial content on one’s ability to fuse imagery. Different types of spatial content (e.g., spatial clutter of man-made structures like buildings, types of trees in a forest, different species of grain in a farm, pixels containing water, plumes/clouds, etc.) and spatial artifacts such as edges and corners are explored as to how they affect the sharpening process. This is achieved by comparing the reference hyperspectral images with fused images, using global metrics (e.g., spectral angle mapper, ERGAS, peak signal-to-noise ratio, cross-correlation, RMSE) as well as performing tasks such as ACE target detection and pixel classification on the reference and fused images and comparing their performance. We explore various ways to quantify these spatial content effects on sharpening and develop different ways of improving fusion algorithms. At time of writing, most fusion algorithms treat the entire scene in the same way. We propose a fusion methodology that adaptively sharpens/spatially enhances hyperspectral imagery according to spatial content. Most of this work deals with the extension of the NNDiffuse algorithm from being a pansharpening algorithm to being a hyperspectral-multispectral fusion algorithm. However, we also explore different unmixing-based algorithms (such as CNMF) as well as deep learning-based algorithms to use in conjunction with NNDiffuse in adaptive/selective image fusion. We also discuss work that has already been done and published in conferences and journals.
Imaging Science (Ph.D.)
Department, Program, or Center
Chester F. Carlson Center for Imaging Science
College of Science
David W. Messinger
Ducay, Rey, "Exploitation of spatial content for enhancing pansharpening and image fusion performance" (2023). Thesis. Rochester Institute of Technology. Accessed from
RIT – Main Campus