The Use of AI (Artificial Intelligence) in Target Classification,
Abstract
A traditional approach to tactical target classification utilizes statistical pattern recognition techniques to classify targets segmented as single objects. Classification of tanks at four to ten kilometers in FLIR imagery is a typical example. There are two areas in which the traditional methodologies fall short. One is that these methods do not readily generalize in order to recognize complexes of many structures, such as missile sites and power plants. The second is that, even within the domain of single blob target classification, it is difficult to incorporate contextual information, for example, the fact that tanks do not appear in the sky, into the statistical pattern recognition approach. Obviously, an ideal target classification system can account both for recognition of complex objects and system methodologies that readily utilize contextual information. Advanced applications for these capabilities include high-value target recognition, passive terminal homing systems, sentry robots, bomb damage analysis, intelligent remote surveillance, landmark based guidance, and autonomous (and semi-autonomous) tactical vehicles. This paper focuses on the first area of generalized (multi-object) target classification. Artificial Intelligence methods provide an approach to target classification that uses knowledge representation and manipulation techniques to extend the base given by statistical pattern recognition. (Author)
Document Details
- Document Type
- Technical Report
- Publication Date
- Jan 31, 1984
- Accession Number
- ADP003022
Entities
People
- D. Panda
- Rahul Aggarwal
- T. Levitt
Organizations
- Honeywell International, Inc.