A Parallel Classification Model for Marine Mammal Sounds Based on Multi-Dimensional Feature Extraction and Data...

Area of Science:

• Marine Biology
• Acoustics
• Machine Learning

Background:

• Deep-sea exploration relies on acoustic signals due to poor visibility.
• The increasing volume of underwater acoustic data necessitates automated analysis methods.
• Developing efficient marine-mammal classification algorithms is crucial for behavioral studies.

Purpose of the Study:

• To design and evaluate an advanced classification model for marine mammal acoustic signals.
• To leverage multi-channel processing and transfer learning for improved accuracy and efficiency.
• To address the challenges posed by large-scale underwater acoustic data.

Main Methods:

• A multi-channel parallel classification model was developed to process multi-dimensional acoustic features.
• Transfer learning was employed to accelerate model convergence.
• Data augmentation techniques were integrated to enhance classification accuracy.
• K-fold cross-validation was used for robust model evaluation.

Main Results:

• The model achieved a mean accuracy of 95.21% in marine mammal classification.
• The model demonstrated high robustness with a standard deviation of 0.65%.
• Consistent performance was observed across multiple evaluation tests.

Conclusions:

• The proposed multi-channel parallel model offers an effective solution for marine mammal acoustic classification.
• Transfer learning and data augmentation significantly improve model performance.
• The model's high accuracy and robustness are vital for advancing deep-sea ecological research.