Characterization of sonar target data using Gabor wavelet features

We present a method of characterization of active sonar target response that makes use of physics-driven Gabor dictionaries. Key innovation is the combination of the useful characteristics of the Gabor wavelet (time/frequency localization) with an empirical approach to dictionary selection for underwater acoustics. We demonstrate over experimental field data that these features are both visually distinct and ideal for separation and classification via support vector machine.