Augmenting Aquaculture Efficiency through Involutional Neural Networks and Self-Attention for Oplegnathus Punctatus...

Area of Science:

• Aquatic Ecology
• Bioacoustics
• Machine Learning

Background:

• Understanding aquatic animal feeding is vital for aquaculture and ecosystem health.
• Current methods for analyzing feeding behavior can be limited.
• Bioacoustic analysis offers a non-invasive approach to study aquatic life.

Purpose of the Study:

• To develop and validate a novel framework for analyzing fish feeding behavior using acoustic data.
• To enhance the accuracy of classifying different feeding intensities in aquatic animals.
• To provide insights for optimizing aquaculture practices and ecosystem management.

Main Methods:

• Audio waveforms were transformed into Log Mel Spectrograms.
• A fusion of features (Discrete Wavelet Transform, Gabor filter, Local Binary Pattern, Laplacian High Pass Filter) was extracted.
• An Involutional Neural Network (INN)-based deep learning model was employed for classification.

Main Results:

• The proposed framework achieved up to 97% accuracy in classifying fish feeding intensities.
• The methodology effectively distinguished various forms of fish feeding behavior.
• The study successfully applied the framework to *Oplegnathus punctatus*.

Conclusions:

• The novel fusion of acoustic features and deep learning provides an effective method for analyzing fish feeding behavior.
• Accurate classification of feeding intensities supports advancements in aquaculture optimization.
• This approach contributes to sustainable marine resource management through enhanced ecological understanding.