Generalized Diagonal Slices and Their Applications in Feature Extraction of Underwater Targets

Abstract

In order to improve the classification accuracy of the radiated noise from underwater targets, generalized diagonal slices and their spectra of third cumulant are proposed by introducing an amendment quantity q into the original diagonal slice. The amendment quantity q represents the distance between the original diagonal slice and other slices parallel to it in bifrequency plane. And the generalized definition degrades into the original one when the amendment quantity q is set to zero. The physical meaning of the generalized diagonal slices spectra depicts lines parallel to the original diagonal slice. And two feature vectors are formed by summation or maximization of every generalized slice. It is validated by simulation that the target features in the bifrequency domain is drastically extracted by summation scheme and that the classification accuracy of 100% for feature vector obtained by summation scheme is steadily resulted by One-against-One (OAO) method of multi classification of Support Vector Machine (SVM).