A pre-attentive vision model for automated data exploration applied to space plasma remote sensing data Ivan Andreevich Galkin As research instruments of large information capacities become a reality, automated systems for intelligent data analysis become a necessity. Scientific archives containing huge volumes of data preclude manual manipulation or intervention and require automated exploration and mining that can at least pre-classify information in categories. The large dataset from the radio plasma imager (RPI) instrument onboard the IMAGE satellite shows a critical need for such exploration in order to identify and archive features of interest in the volumes of visual information. In this research we have developed such a pre-classifier through a model of pre-attentive vision capable of detecting and extracting traces of echoes from the RPI plasmagrams. The overall design of our model complies with MarrÕs paradigm of vision where elements of increasing perceptual strength are built bottom up under the Gestalt constraints of good continuation and smoothness. The specifics of the RPI data, however, demanded extension of this paradigm to achieve greater robustness for signature analysis. Our pre-attentive model now employs a feedback neural network that refines alignment of the oriented edge elements (edgels) detected in the plasmagram image by subjecting them to collective global-scale optimization. The level of interaction between the oriented edgels is determined by their distance and mutual orientation in accordance with the Yen and Finkel model of the striate cortex that encompasses findings in psychophysical studies of human vision. An additional effort has been made to pre-process the raw image data to eliminate noise and to detect, interpret and remove the resonance signatures that resulted from the local response of the plasma to the RPI signal. The developed models have been implemented in an operational system ÒCORPRALÓ (Cognitive Online Rpi Plasmagram Ranking Algorithm) that currently scans daily submissions of the RPI plasmagrams for the presence of echo traces. Qualifying plasmagrams are tagged in the mission database, making them available for a variety of queries. An analysis of the performance of CORPRAL is given and directions for future research are outlined. _______________ Ph.D. Thesis, University of Massachusetts, Lowell, 2004