Construction of an AI-based poultry health monitoring device

The increasing demand for poultry production has underscored the necessity for automated and objective systems capable of continuously monitoring animal behavior and health. Traditional observation-based methods are limited by subjectivity, high labor requirements, and poor scalability in modern intensive farming. This study presents the design, development, and validation of an IoT-enabled poultry behavior monitoring device employing machine learning (ML) and multi-modal sensor integration for real-time health classification. The system incorporates an ESP32-CAM microcontroller, Inertial Measurement Unit (IMU), and environmental sensors (temperature, humidity, gas, and light) to capture and process synchronized behavioral and environmental data. A neural network model, trained using Edge Impulse Studio and deployed at the edge, achieved an overall validation accuracy of 88.6% with an AUC of 0.98, effectively classifying health states including Healthy, Coccidiosis, and Salmonellosis. Feature importance analysis revealed IMU-derived motion data and air quality as primary indicators of health anomalies, validating the ethological and environmental frameworks. The proposed system demonstrates high potential for real-time, low-cost, and scalable precision livestock monitoring, offering early disease detection and enhanced welfare management within poultry operations.