Automatic detection of marine mammals using information entropy

Area of Science:

• Acoustics
• Marine Biology
• Signal Processing

Background:

• Passive acoustic monitoring (PAM) is crucial for marine mammal research.
• Increasing data volumes in PAM studies necessitate efficient automated detection methods.
• Detecting diverse vocalizations across multiple species remains a significant challenge.

Purpose of the Study:

• To develop and evaluate an automatic detector for marine mammal vocalizations.
• To address the challenge of detecting a wide variety of calls from different marine mammal species.
• To improve the efficiency of signal analysis in large acoustic datasets.

Main Methods:

• Utilized spectral entropy as a measure of information content in acoustic signals.
• Developed an automatic detector based on spectral entropy.
• Compared the performance of the entropy detector against peak-energy detectors using receiver operating characteristic (ROC) curves.
• Tested the detector on underwater recordings from the Western Canadian Arctic.

Main Results:

• The spectral entropy detector demonstrated superior performance compared to peak-energy detectors.
• The detector operated considerably faster than real-time.
• Successfully detected vocalizations from various cetacean and pinniped species.
• The method proved effective for analyzing large volumes of acoustic data.

Conclusions:

• Spectral entropy provides an effective basis for an automatic marine mammal vocalization detector.
• The developed detector is efficient and suitable as a preliminary step in automated signal analysis.
• This method can significantly aid in processing large datasets from passive acoustic monitoring.
• Further integration with classification algorithms is recommended for comprehensive signal identification.