The Remote Sensing Hyperspectral Engine (RSHE) is a special-purpose, portable computer that performs high-performance processing of hyperspectral image data collected by a remote-sensing optoelectronic apparatus. Typically, the remote-sensing apparatus is airborne or spaceborne, the images are of terrain, and the purpose of collecting and analyzing the image data is to estimate the spatially varying abundances of materials of interest. Remote-sensing applications in which the RSHE could prove beneficial include assessment of crops, exploration for minerals, identification of military targets, urban-planning studies, environmental assessment, and large-area search-and- rescue operations.

This system is designed to perform the overall functions of (1) extracting a spectral signature of each pixel from the data for a given image and (2) processing the spectral signatures of the pixels to unmix the superimposed spectra of different materials and thereby obtain estimates of the abundances of materials in each pixel of the image. What distinguishes this system from other such systems are the specifics of its implementation in hardware and software. The hardware comprises mostly commercial off-the-shelf modules and assemblies chosen to afford the required computational capabilities while fitting in an ultracompact package. The most notable aspect of this system is software that processes hyperspectral image data robustly and efficiently and provides enhanced means for displaying and otherwise using the results of processing to facilitate understanding of images.

The software performs so many advanced functions that it must suffice to list only a few of them here:

  • The hyperspectral image data are processed by using a Universal Robust Filtering (URF) Software Package (a commercial product developed by the American GNC Corp.) coupled with an optimization algorithm to iteratively search through the space of candidate solutions. Because an optimization algorithm that starts from a tabula rasa involves many iterations and thus is inherently time-consuming, this system takes advantage of the existing correlation between neighboring pixels to reduce the amount of search needed. After obtaining the solution for the first pixel by using the optimization algorithm, the solutions for the rest of the pixels are derived from the computationally efficient robust filtering algorithm.
  • The robust nonlinear spatial filter includes the applicable measurement equations and an extended Kalman filter. Constraints on the abundances of the materials of interest, expressed through mathematical models, are incorporated into the filter. Thus, the abundance-unmixing problem is transformed into an augmented, nonlinear filtering problem that is solvable by use of nonlinear, extended Kalman filtering techniques.
  • Each material of interest is represented in a reference library of spectral signatures that have been processed into orthonormal basis vectors.
  • Identification of materials of interest involves the use of a fuzzy neural network to correlate preprocessed hyperspectral image data with orthonormal reference-signature vectors.
  • The graphical user interface facilitates the initiation of processing, retrieval of images, and display of image data. For example, one can choose N monochrome two-dimensional displays for N materials of interest; the brightness of each pixel of one of the images is an indication of the abundance of the material in question at the pixel location.

This work was done by American GNC Corp. for Stennis Space Center.

In accordance with Public Law 96-517, the contractor has elected to retain title to this invention. Inquiries concerning rights for its commercial use should be addressed to

American GNC Corp.
888 Easy Street
Simi Valley, CA 93065

Refer to SSC-00138, volume and number of this NASA Tech Briefs issue, and the page number.

Photonics Tech Briefs Magazine

This article first appeared in the January, 2002 issue of Photonics Tech Briefs Magazine.

Read more articles from the archives here.