Pattern Recognition Techniques for Radar Aircraft Identification.

摘要

Some problems concerning the design and evaluation of a pattern recognition system for low-frequency radar aircraft identification are investigated. A decision-theoretic model that includes a model of noise perturbations of the feature-vector, and a model of uncertainity associated with the measurement of parameter values has been developed. The performance of the minimum-risk classifier based on the decision-theoretic model is compared to those of the nearest-neighbor classifiers. This comparison is made in terms of the following measures: misclassification probability, classification bias, computational efficiency and robustness. A feature-extraction algorithm based on the decision-theoretic model is developed. Alternative procedures for the problem where multiple radar scans are available have been compared. These procedures include a sequential procedure based on the decision-theoretic model, and a procedure based on taking a majority-vote over multiple decisions of a given classifier. (Author)