Automatic Target Recognition: Some Considerations

Machine analysis of aerial reconnaissance data is desirable. New acquisition techniques, equipment, and sensors, plus a growing demand for rapidly obtained, accurate information necessitate machine performance of tasks presently accomplished by observers and interpreters. Many of these tasks can be described as pattern recognition or information processing functions. This paper discusses significant problems encountered when attempting to identify objects or regions automatically from aerial photography. Problems include sensor distortions, variations in target environment, and concepts of design and "teaching" of an actual device. Operational requirements such as processing speed and flexibility restrict the types of solutions available. Approaches to target recognition with relative merits and shortcomings are presented; these include shape correlation, feature extraction, and image quantization. The effect of mission requirements on system parameters is discussed. A bibliography is included.