Classifying multi-frequency fisheries acoustic data using a robust probabilistic classification technique.

Area of Science:

• Fisheries Science
• Acoustic Data Analysis
• Statistical Modeling

Background:

• Fisheries acoustic data analysis is crucial for stock assessment.
• Existing methods may lack robustness in diverse or poorly understood systems.
• Probabilistic classification offers a powerful approach to interpreting complex acoustic signals.

Purpose of the Study:

• To develop and demonstrate a robust probabilistic classification technique for multi-frequency fisheries acoustic data.
• To utilize expectation maximization of finite mixture models for sample classification.
• To validate the technique in both low-diversity and species-rich marine environments.

Main Methods:

• Employed expectation maximization of finite mixture models for probabilistic classification.
• Utilized the Bayesian Information Criterion for optimal cluster number selection.
• Classified acoustic samples based on probabilities of cluster membership.

Main Results:

• Successfully classified acoustic samples using the developed probabilistic technique.
• Demonstrated the method's utility in the well-known Gulf of Alaska system.
• Showcased the technique's effectiveness in the species-rich, less-explored Mid-Atlantic Ridge system.

Conclusions:

• The probabilistic classification technique provides a robust method for analyzing fisheries acoustic data.
• The approach is applicable to diverse marine ecosystems, from well-known to poorly understood.
• This method enhances the ability to classify and understand fish populations using acoustic data.