Recognition Method for Broiler Sound Signals Based on Multi-Domain Sound Features and Classification Model.

Area of Science:

• Animal Science
• Bioacoustics
• Machine Learning

Background:

• Existing broiler sound classification research has limitations in feature extraction, analysis, and model selection.
• There's a need for a more robust method to accurately identify broiler vocalizations for improved animal welfare and management.

Purpose of the Study:

• To propose and validate a novel recognition method for broiler sound signals.
• To enhance the accuracy and reliability of broiler sound classification through advanced feature engineering and model optimization.

Main Methods:

• Collected and filtered broiler sound signals, extracting 60 multi-domain features (time, frequency, MFCC, sparse representation).
• Applied feature selection, reducing features to 30, and trained seven classification models, optimizing a k-Nearest Neighbor (kNN) model.
• Utilized min-max standardization and majority voting for enhanced prediction accuracy.

Main Results:

• Feature selection improved classification accuracy by 3.1%, and kNN parameter optimization by 1.2%.
• The optimized kNN model achieved a highest classification accuracy of 94.16% during training.
• The final model demonstrated 93.57% classification accuracy and 99.12% recognition accuracy on test data.

Conclusions:

• The proposed method effectively addresses limitations in existing broiler sound recognition research.
• Multi-domain feature extraction, selection, and optimized kNN classification significantly enhance recognition accuracy.
• This approach offers a promising tool for precise broiler sound analysis in practical applications.