SeqFusionNet: A hybrid model for sequence-aware and globally integrated acoustic representation.

Area of Science:

• Bioacoustics
• Machine Learning
• Wildlife Conservation

Background:

• Animal vocalizations are crucial for communication, species conservation, and biodiversity tracking.
• Traditional visual monitoring methods have limitations in distance and environmental conditions.
• Call-based monitoring offers a more effective and direct approach to studying animals.

Purpose of the Study:

• To introduce SeqFusionNet, an advanced animal sound classification model.
• To evaluate the model's performance across diverse acoustic datasets.
• To demonstrate the model's robustness and generalization capabilities for bioacoustic monitoring.

Main Methods:

• Developed SeqFusionNet, integrating Transformer for sequential encoding and MLP for global perception.
• Compiled and utilized four acoustic datasets: pig, bird, urban sound, and marine mammal.
• Conducted extensive experiments to assess cross-species applicability and recognition accuracy.

Main Results:

• SeqFusionNet achieved high accuracy in classifying various animal calls: 95.00% for pigs, up to 95.24% for birds, 97.50% for marine mammals, and 94.39% for urban sounds.
• The model demonstrated superior performance compared to existing methods.
• SeqFusionNet showed strong robustness and generalization across different species and datasets.

Conclusions:

• SeqFusionNet is an effective framework for automated bioacoustic monitoring.
• The model supports wildlife preservation, ecological studies, and environmental sound analysis.
• The research provides an expandable and efficient solution for analyzing animal vocalizations.