Research on Evaluating the Filtering Method for Broiler Sound Signal from Multiple Perspectives

Area of Science:

• Agricultural Engineering
• Acoustics
• Signal Processing

Background:

• Broiler vocalizations offer insights into animal well-being and health.
• Farm environments introduce significant noise into audio recordings, complicating signal analysis.
• Effective noise reduction is crucial for accurate interpretation of broiler sounds.

Purpose of the Study:

• To evaluate and compare various filtering techniques for broiler sound signals.
• To identify the optimal filtering method for enhancing signal quality and subsequent analysis.
• To assess filtering performance from both signal integrity and classification accuracy perspectives.

Main Methods:

• Applied four filtering methods: Basic Spectral Subtraction, Improved Spectral Subtraction, Wiener filtering, and Sparse Decomposition.
• Evaluated methods using signal-to-noise ratio (SNR) and root mean square error (RMSE) for signal quality.
• Assessed classification accuracy using kNN and Random Forest on filtered datasets.

Main Results:

• Improved Spectral Subtraction yielded the highest SNR and lowest RMSE in signal-level analysis.
• Wiener filtering enabled the highest prediction accuracy (88.83% with kNN, 88.69% with Random Forest) in recognition-level analysis.
• Wiener filtering demonstrated superior performance in overall signal and recognition quality.

Conclusions:

• Wiener filtering is identified as the most effective method for broiler sound signal processing.
• This research provides a foundation for health monitoring through analysis of high-quality broiler sounds.
• Findings are applicable to noise reduction in sound signals from various livestock and poultry.