Machine Intelligence Applied to Radar Object Modeling
MASSACHUSETTS INST OF TECH LEXINGTON LINCOLN LAB
Pagination or Media Count:
In this report, we discuss a machine intelligence approach to modeling simple space objects from radar range-Doppler images. The data is multidimensional in nature with additive noise, distortion, and missing points. The relevant features to be extracted from the radar data include the position of all scattering centers in body coordinates, identification of the scattering center type sphere, corner, edge, etc., and motion parameters for the object. Our goal is to produce a representation of the imaged 3-dimensional object that is appropriate for recognizing the object as an example of something we have seen before and cataloged, recognizing the object as an uncataloged object, or determining discrepancies between the recognized object and our expectations of its appearance. We have built a recognition system with three major conceptual modules. The first of these is a set of signal processing primitives that are directed at the data to select subsets of data, extract features, and compare extracted features with the data to produce confidence measures. The second major module is the semantic model building and matching scheme. This component takes the data-derived features and produces a semantic model which is then matched against a catalog of stored semantic models for object identification.
- Active and Passive Radar Detection and Equipment