A novel method of underwater multitarget classification based on Multidimensional Scaling analysis

Abstract

In order to solve the problem of robustly classifying underwater multiple targets in shallow sea, a novel classification method based on Multidimensional Scaling (MDS) is proposed. This algorithm extracts the robust and distinct feature difference between targets by means of MDS, and optimizes the feature distance by combining with kernel function. A modified K-means classifier is utilized to cluster the extracted features without knowing the prior information of class number. Experiment results on real sonar detecting data indicate that the classifying probability increases by 13.4% compared with PCA, and the probability and robustness of underwater target classification are improved effectively.