Mallat's Matching Pursuit of sperm whale clicks in real-time using Daubechies 15 wavelets

Abstract

This article presents how the wavelet matching pursuit (MP) algorithm can perform in real-time, ocean noise elimination and whale clicks detection and segregation. This work is based on two recent advances: the first [Adam 05]proposes to analyze sperm whale signals with Daubechies 15 wavelets. The second is the fast implementation of the MP by INRIA Rennes team [MPTK 06]. We first show that one can interface DB15 with this toolkit. Second, we demonstrate, for the first time, a real use of DB15 on two long whale records reference datasets containing one or several sperm whales to address the click labeling issue. Experiments are conducted on AUTEC Bahamas data, whales are from 5 to 10 km far from the bottom mounted hydrophone. We show that DB15 are efficient for real time click detection, ocean noise and echoes removal. We get 85% of click detection for less than 20% of false alarm. We finally apply our approach to 3 co-clicking whales, and we show promising results of automatic fast segregation and click labeling of simultaneous emitting whales.