Logan Smith with Julia Dobrosotkaya

Analysis and processing of hyperspectral data with the purpose of classification

Hyperspectral imaging is an advanced imaging technique that measures visible and near-infrared light reflecting off a surface. Hyperspectral imagery has a wide range of applications from geospatial sciences to ecology, surveillance and more. A hyperspectral image is a 3D structure with a spectrum of values associated with every pixel corresponding to the image intensity at a fixed spatial location recorded at different wavelengths. These spectra can be compared to known materials, and then classified.

Hyperspectral data processing has been given a lot of attention during the past decade, but the problem of classification is still open. We intend to perform an in-depth study of the existing classification and dimension reduction methods for hyperspectral data. We will attempt to improve those by integrating the most successful existing methods into a joint framework and/or adjusting the methodology for a specific subclass of data with the goal of improving the performance in either the quality of output or the computational efficiency. We will use the publicly available hyperspectral datasets  and spectral signatures (SpecTIR® Remote Sensing Division website, NASA).