Nonlinear classifier combination for a maritime target recognition task

This research proposes a nonlinear combination of an ensemble of classifiers to improve pattern recognition performance. A maritime target recognition application is considered. A database of radar range profiles with six ship classes from various aspect angles were created. Five structurally based features are defined on the simulated range profiles. Three kinds of classifiers are used: neural network, Bayes, and nearest neighbor. The proposed nonlinear combination scheme utilizes a neural network combiner. The performance of this combiner is compared to individual classifiers as well as two other combination approaches.