Application of time-frequency and time-scale analysis to underwater acoustic transients

The authors discuss an ongoing program to find algorithmic analysis techniques that can extract features of underwater acoustic transients for purposes of high-confidence classification. It has been found that no one analysis technique can adequately capture features from all the possible transient classes in a large database of actual transients. Side-by-side comparisons of numerous time-frequency and time-scale representations of an unclassified acoustic transients are presented. The techniques illustrated include the short-time Fourier transform (STFT) tuned to different analysis bandwidths, the Gabor-wavelet time-scale representation (TSR), the double diamond time-frequency representation (TFR), the adaptive optimal radially Gaussian kernel (AOK) TFR, and model-based TFRs. The strengths and weaknesses of the various analysis representations are discussed