AISRP workshop '97 presentation,    E. Merényi, U of Arizona, LPL

2. The Vision and Objectives

The long term goal, that goes beyond this project: A small, lightweight, massively parallel hardware module on-board

• with massively parallel classification algorithms implemented in it;
  • These must be sophisticated and robust enough to efficiently exploit hyperspectral images, i.e. to extract all scientifically relevant knowledge that is contained in the data.

• with extremely high processing speed (classify several 100K hyperspectral pixels per second), on-chip learning capability, good scale up properties without loss of speed, and adequate I/O speed

The objective of this project is to develop the algorithmic (software) capability under the first bullet above. Currently, analysis techniques, and notably classifiers, for high-dimensionality data (more than a dozen or so image bands) lag severely behind the powers of advanced sensors in detecting intricate details.

The combined capabilities under the two bullets will be significant for on-ground data analyses too, in contexts such as the Earth Observing System. An AVIRIS-class image will be classified in seconds, into a large number of classes.