Jiarui Yi, The University of Texas at El Paso; Miguel Velez-Reyes, The University of Texas at El Paso; Hector Erives, The University of Texas at El Paso
Keywords: hyperspectral remote sensing, hyperspectral unmixing, non-resolved object characterization
Abstract:
Space assets are critical for critical for USA defense, security and economic wealth. Remote sensing is an important technology to gain situational awareness of the environment surrounding space assets. Ground-based space telescope technology cannot spatially resolve objects in space that are distant (orbits beyond 1,000 km altitude, e.g. GEO) or that are small (e.g. CubeSats). These objects are denoted as non-resolved space objects. Hyperspectral remote sensing has been proposed as a technology to extract quantitative information about non-resolved space objects. The high spectral resolution of hyperspectral sensors contains information about the material composition of the non-resolved object from materials contribution to the measured spectra. Even though the object cannot be spatially resolved, it may be spectrally resolved.
Hyperspectral unmixing is a technique used to decompose mixed measured spectral signatures into the spectral signatures of constituent materials and their abundances. In terrestrial applications, unmixing has been widely studied looking at images that contain spectral and spatial information of the object of interest. In the case of non-resolved space objects, the authors have proposed the use of the spectro-temporal signature of temporal traces collected while the space object is in transit in the field of view of the hyperspectral sensor to extract material composition information. A big challenge for this approach is that the collected spectro-temporal signature may no be rich enough to extract the material composition using blind hyperspectral unmixing methods.
In this paper, we present a simulation study that shows how prior information about space object composition may be useful in helping dealing with the lack of richness of spectro-temporal signatures. We study the use of the constrained non-negative factorization with partial endmembers knowledge to address this problem. We look at performance as a function of the resolution and the similarity of endmembers. A simple simulation model of an object rotating over a background is used in the study. We will also show preliminary results using the Digital Imaging and Remote Sensing Image Generation model (DIRSIGTM). This is a physics-based, image and data simulation model used to generate synthetic imagery.
Date of Conference: September 14-17, 2021
Track: Non-Resolved Object Characterization