Feature extraction of infinite elastic cylindrical objects underwater

Effective feature extraction is fundamental for the classification of underwater targets. This paper studies the methods of feature extraction of infinite elastic cylindrical targets on the basis of theoretical analysis of the form function. Through investigating how physical parameters and thickness influence the resonances in the form functions of infinite elastic cylindrical objects, the interval between two adjacent resonances of the same type and the Q values of those resonances are extracted, and the linear relation between the resonances of the same type and their Q values is found out. Thus, a three-dimensional feature vector, which is composed of the interval between two adjacent resonances of the same type, the variation of Q values of those resonances and the number of the resonance types, is proposed. Furthermore, the clustering of the cylindrical targets with different physical properties and thicknesses is conducted based on the proposed feature vector, which is calculated from the simulated backscattering signals. It demonstrates that the feature vector is valid in separating metal cylinders from nonmetal ones, or classifying cylindrical shells with different thicknesses. Copyright ? (2014) by the International Institute of Acoustics & Vibration All rights reserved.EI