Multimodal Sparse Time–Frequency Representation for Underwater Acoustic Signals

Multiple features can be extracted from time-frequency representation of signals for the purpose acoustic event detection. However, many underwater are formed by multiple events (impulsive and tonal), which generates difficulty on high-resolution TFR each component. For characterization such different events, we propose an anisotropic chirplet transform to achieve with high energy concentration. Such applies a varying Gaussian window compensate component while suppressing unwanted noise. Using set directional ridges obtained TFR, structure-split-merge algorithm is designed reconstruct multimodal sparse representation, provides instantaneous frequency time features. Specifically, pulsed-to-tonal ratio, based these features, computed distinguish two signals. The presented method validated using shallow water experimental communication large sequences harmonics pulsed bursts common whales.