A multi-sensor target recognizer (MSTR)

Abstract

The problem of designing an MSTR with an optimal fusion center is addressed. Since it was determined that signal processing and classification are best performed at the sensors, the MSTR described is constructed with multiple sensor classifiers; each sensor classifier is designed with some optimal recognition scheme and classifies targets independently of other sensor classifiers. The result of target recognition by an individual sensor is transmitted to a data fusion center that has been optimally designed. The MSTR design is illustrated using radar and infrared (IR) sensors. A specific design example for a two-sensor, three-class MSTR with Gaussian data showed a 14% improvement in the average probability of correct classification (P/sub cc/) over a single-sensor system. This design was further demonstrated in a radar-IR MSTR using field radar and field FLIR (forward-looking infrared) data. The performance results show an average 12% P/sub cc/ improvement over radar alone and 9% P/sub cc/ improvement over IR alone.<>