Automated Poultry Health Monitoring through Acoustic Analysis Using Convolutional Neural Networks

Improving animal welfare and reducing losses in poultry breeding and production systems hinge on the early detection and warning of contagious diseases among chickens. Traditional methods for controlling and diagnosing poultry diseases often fall short, leading to significant mortality and decreased output. This study presents an automated poultry health management algorithm based on Convolution Neural Networks (CNNs) to identify healthy and unhealthy chickens using acoustic analysis of their vocalizations. The proposed approach leverages Mel-Frequency Cepstral Coefficients (MFCC) for feature extraction from audio signals of chickens that exhibit respiratory diseases. The CNN model, which comprises convolution layers, dropout layers, and batch normalization, was trained and evaluated on a dataset of 346 audio signals collected from poultry farms. The results demonstrate high accuracy (94.59%), precision (96%), recall (96%), and F1-score (96%) in classifying healthy and unhealthy chicken sounds, outperforming previous methods. This study underscores the potential of voice-based diagnostic tools in poultry health management, offering prospects for early intervention and enhanced health outcomes.