Abstract

Several requirements distinguish remote sensing applications of pattern classification. The shape of the pattern may not be of primary importance. Often, observed regions (cities, crops, forests, oceans, etc.) display characteristic textures but no particular shape. Even when objects to be recognized have particular shapes, they are often embedded in such large fields of view that too few pixels are available to distinguish shape. An optical processor has been designed that recognizes objects based on their spectral and texture characteristics. A pixel-wide strip of the scene is processed at a time. The spectrum of each pixel of the scene is spread in the direction perpendicular to the strip by a grating. The resulting spatial spectral image that results can drive the input side of an optically addressed spatial light modulator. Conversion of the multispectral information into coherent monochromatic light allows further processing by using optical correlation techniques.

© 1992 Optical Society of America