note 4.1  AISRP workshop '97 presentation,    E. Merényi, U of Arizona, LPL

The functional diagram of the project.  The middle section contains procedures that are massively parallel, and therefore lend themselves to parallel hardware implementation. The Artificial Neural Net block  contains a Self Organizing Kohonen Map for clustering (unsupervised classification), and a supervised ANN classifier. Data can be explored first with an unsupervised classification - visualization - human interaction loop, based on which training pixels may be selected with high confidence. A subsequent supervised classification can also be refined with human feedback until class separations are satifactory (imagine an exploratory first phase of orbital mapping). Then the system can be put on autopilot, classifying incoming data continuously until some condition change occurs (for example, a significant change in the data characteristics, which is constantly evaluated), and human interaction is warranted.