Entropy-Based Automatic Detection of Marine Mammal Tonal Calls

Abstract

Hydrophones are deployed throughout the ocean to perform passive acoustic monitoring. This technique is a powerful tool for marine mammal sound detection due to its advantage of being able to collect data overnight, year-round, and in inclement weather. However, hundreds of terabytes of data produced each year pose a significant challenge for data analysis. The aim of this study was to investigate the use of entropy-based techniques to achieve automatic detection of marine mammal tonal calls in passive acoustic monitoring data. A weighted spectral entropy technique was developed to alleviate the impact of underwater noise along with a novel algorithmic detector. The detector includes an adaptive bandpass filter, a time–frequency domain transform, and a likelihood ratio test for calculating the optimal detection threshold in addition to the Weighted Spectral Entropy Technique. The proposed entropy-based technique and the automatic detector were assessed with synthetic and real-world data and the performance was compared to other state-of-the-art techniques. The results indicate that the proposed method outperforms the other techniques when evaluated using various types of low signal-to-noise ratio tonal signals.