A novel, automatic method for the detection, extraction and characterization of marine mammal tonal calls is presented. Signals are automatically detected from the spectrogram, isolated using region-based segmentation, extracted and finally characterized by means of a fixed number of radial basis function (RBF) coefficients. A total of sixteen RBF coefficients are sufficient to accurately capture the time–frequency information contained in the calls. These coefficients can be later used to classify signals based on their characteristics. New specific functions for contour extraction and cross-resolution have been developed. The performance of the method has been extensively tested using simulated signals and a set of recordings covering a significant range of situations that can be encountered at sea.
