A PAMGuard catalog for comparing detection and classification performance of two mode decomposition algorithms

Abstract

The ONR project “Application of an Empirical Mode Decomposition (EMD) detection and classification process to environments for Naval monitoring and detection” created empirical and variational mode decomposition analysis workflows to detect and cluster diverse underwater signals across four distinct datasets. Twenty-five call types from the Bering and Chukchi Seas, Aeon’s North Atlantic sites, CTBTO hydroacoustic stations, and the Chambal River were detected and clustered with varying levels of success. The final precision, recall, and accuracy results of these call types will be presented. Evaluating the performance of our developed detection and classification algorithms involved a comparison with the widely used open-source software, PAMGuard. However, obtaining information about the parameters used to construct detectors for different call types in PAMGuard proved challenging due to limited literature availability. To the best of our knowledge, some call types had never been processed in PAMGuard, and for those that had, not all parameters required for adapting the detectors to our datasets were consistently published. To facilitate the use of our PAMGuard detectors by other researchers for their own datasets, we will present a user-friendly guide that will outline the features that yielded optimal PAMGuard performance for each call type.