An associative and nonparametric digital information processing technique for application in real-time, on-board pattern recognition

Abstract

An inherent characteristic in a synoptic data acquisition system, such as Landsat, is a voluminous data rate. A proposed method of data reduction or data compression is to perform real-time discriminant analysis on the acquired data aboard the data collecting platform. Such a data analysis system, however, requires both high computational speed and minimal computational hardware. The pattern recognition algorithms that are currently applied to the remote sensing data do not collectively meet those special requirements. As a feasible solution to the problem of on-board classification of the Landsat MSS data, the Real-time, Associative Pattern Identification (RAPID) technique is developed and described herein. The RAPID technique is formulated based on the criteria of minimizing (1) probability of misclassification, (2) computational time and (3) computational hardware. In part, these constraints are satisfied with a nonparametric approach and assumption of statistical independence of the components of the feature vector. Further optimization of the classification technique is achieved by design of a special purpose digital processor which is configured around the associative information storage and retrieval concept.