Support Vector Machine Applied to Underwater Target Classification

Underwater target classification is a complex task, due to the difficulty in identifying non-overlapping and stable feature set. It is required to choose the right algorithm, approach and technique, or the best combinations of approaches and techniques from a large set of options available in the literature for the specific problem. A binary classifier can tackle the problem by decomposing multiclass problem into binary class. This paper addresses the multiclass underwater classification problem using binary classifier -- Support Vector Machine (SVM). Three methods "all-against-all," "all-against-all Hierarchical," "one-against-all"("AVA", "AVA-H", "OVA") are tried out and performance using a particular feature derived from real data set is compared. A number of metrics are used to compare the performance. OVA gives a better performance with less computation compared to other methods.