Variational Mode Decomposition-Based Identification of Pygmy Blue Whales Song Units

• Published in: IEEE sensors journal Vol. 24; no. 11; pp. 17963 - 17973
• Main Authors: Liang, Yue, Al-Badrawi, Mahdi H., Seger, Kerri D., Kirsch, Nicholas J.
• Format: Journal Article
• Language: English
• Published: New York IEEE 2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Adaptive noise modeling; Background noise; Classification; Datasets; Decomposition; Electronic mail; Feature extraction; Marine mammals; Noise; Noise generation; Population (statistical); Population density; pygmy blue whale; sensor data processing; Sensors; Sociology; Sound sources; variational mode decomposition (VMD); Whales
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2024.3388330

Detection and classification of cetacean vocalizations in acoustic datasets play critical roles in understanding the impacts of various human-made sound sources on cetaceans. Knowing migration route locations, feeding success rates, and population densities can help reduce our impacts on their life functions and aids in conservation efforts for these mammals. However, due to the amount of data that need to be processed and the amount of variance that exists in ocean noise, identification of large marine mammals is challenging. The proposed method relies on modeling and generating noise-only copies of a given data sample and uses that noise-only copy to determine whether a signal is contained in the sample or not. With the aid of this noise modeler, a variational mode decomposition (VMD) technique is then applied to detect signals and extract their features. Finally, the extracted features are used to group detected signals into four different types of blue whale units, namely, Units 1-3 of the Sri Lankan pygmy blue whale song and the Diego Garcia downsweep (Chagos) song, which is thought to be from a different pygmy blue whale population. The evaluation of statistical indices revealed the efficacy of the proposed system, since overall precision was 99.5%, and recall was 87%. With this automated detection/classification system that adjusts for changing background noise conditions, analyzing the many datasets containing multiple blue whale songs can be more efficiently accomplished.