1 Feature Extraction from Underwater Signals using Time-Frequency Warping Operators Cornel IOANA * , Member IEEE, André QUINQUIS * , Senior Member IEEE Yann STEPHAN ** , Member IEEE * : ENSIETA, E3I2 Research Centre, 2 rue François Verny, 29806 Brest, France, Phone : +33.298.34.88.86, Fax : +33.298.34.87.50 Emails : ioanaco@ensieta.fr, quinquis@ensieta.fr ** : EPSHOM, Military Center of Oceanography, 13 rue de Chatellier, 29603 Brest Phone : +33.298.22.10.85, Fax : +33.298.22.18.64 Email : yann.stephan@shom.fr Abstract – Processing marine-mammal signals for passive oceanic acoustic tomography or species classification and monitoring are problems that have recently attracted attention in scientific literature. For these purposes, it is necessary to use a method which could be able to extract the useful information about the processed data, knowing that the underwater environment is highly non-stationary. In this context, time- frequency or time-scale methods constitute a potential approach. Practically, it has been observed that the majority of time-frequency structures of the marine- mammal signals are highly non-linear. This fact affects dramatically the performances achieved by the Cohen's class methods, these methods being efficient in the presence of linear time-frequency structures. Fortunately, thanks to the warping operator principle, it is possible to generate other class of time- frequency representations (TFRs). The new TFRs may analyze non-linear chirp signals better than Cohen's class does. In spite of its mathematical elegance, this principle is limited in real applications by two major elements. First, as we will see, its implementation leads to a considerable growth of the signal length. Consequently, from operational point of view, this principle is limited to short synthetic signals. Secondly, the design of a single warping operator can be inappropriate if the analyzed signal is multi-component. Furthermore, the choice of “adapted” warping operator becomes a problem when the signal components have different time-frequency behaviors. In this paper, we propose a processing method of marine-mammal signals, well adapted to a real passive underwater context. The method tries to overcome the two limitations mentioned before. Also, the first step consists in data size reducing by the detection of the time-frequency regions of interests (ROIs). Furthermore, in each ROI, a technique which combines some typical warping operators is used. The result is an analytical characterization of the instantaneous frequency laws of signal components. The simulations on real underwater data show the performances of this method in comparison with classical ones. Keywords: time-frequency analysis, wavelet transform, high-order statistics, underwater environment, warping operators, passive tomography 1. INTRODUCTION Motivation for processing marine-mammal signals stems from increasing interest in the behavior of endangered marine mammals, reflected in a number of publications in the scientific literature [1,2]. The ultimate goal of the current research in this field is to develop tools for the simultaneous localization of mammals and analysis of the emitted signal for species identification and monitoring. On the other hand, the characterization of underwater environment is a challenging topic, due to the richness of the potential information that can be extracted for navigation or communication, for example. One of the major methods is the oceanic active tomography [3], which provides an environmental characterization using a man-made transmitted signal. Nevertheless, it is possible to hal-00317991, version 1 - 3 Sep 2008 Author manuscript, published in "IEEE Journal of Oceanic Engineering 31, 3 (2006) 628-646" DOI : 10.1109/JOE.2006.875275