GA based selection and parameter optimization for an SVM based underwater target classifier

Underwater target classification is a very demanding task owing to ever changing complicated nature of the underwater communication channels. Underwater target classification system identifies targets from a mixture of underwater events by its characteristic signature. The characteristic signatures pertaining to each target are patterned by feature recognition algorithms operating on hydrophone captured data. In this paper, an SVM target classifier is used to distinguish between targets of 4 acoustic classes. The performance of the classifier is improved by automating the selection of optimal algorithmic parameters. This paper attempts towards optimal selection of SVM parameters, kernel and kernel parameters using genetic algorithm.