AI-based early warning mechanism for poultry farms: Evaluating acoustic signal algorithms for bird health and sustainable production

Abstract

This study examines the rising consumption of animal products in Africa amid concerns about declining production. It suggests that food production needs to increase by 25% to meet the demands of a growing population. The experimental study involved 100 poultry birds divided into two groups: one inoculated with chronic respiratory disease (CRD), and one uninoculated. Over 65 days, audio signals were collected three times daily in a controlled environment with ethical approval. A nano 32 BLE sensor was used to collect a dataset of 346 audio signals from the farm. These signals were categorized as healthy, disease-related, or noisy. To identify the most effective model for early detection of poultry diseases, three algorithms utilizing audio signals were evaluated: MFCC, MFE, and Spectrogram. Results showed the Spectrogram algorithm outperformed others, with 99.2% accuracy, 99.3% F1-score, and 0.05 loss. The MFCC algorithm had 85.6% accuracy, 85% F1-score, and 0.38 loss, while the MFE algorithm achieved 97.4% accuracy, 97.3% F1-score, and 0.08 loss. Implementing it can support sustainable development goals 1 (No Poverty), 2 (Zero Hunger), 3 (Good Health and Well-being), and 12 (Responsible Consumption and Production) by improving poultry farming and reducing economic losses.