Real-Time Infection Detection System in Broiler Farm using MobileNetSSD Model

Abstract

The early detection of diseases in poultry farms is very important in safeguarding flock health and reducing economic losses. Outdated method of monitoring poultry health involves manual examinations, which are time consuming, labor-intensive and prone to inaccuracies. To curtail these challenges, this study presents a real-time infection detection system using lightweight object detection model called MobileNetSSD model for efficient and automated health monitoring. The system consists of deep learning techniques with affordable hardware that support real-time detection, tracking and analysis of broilers movement patterns in farm A, that consist of untagged healthy broilers and Red tagged sick broilers. The exercise was repeated three times to obtain movement threshold 84.9cm for sick broilers and 213.03cm healthy broilers, the outcome produced from reference farm A was used to analyze farm B and farm C. The model achieved 87% average accuracy, 93% average precision 78% average recall and 84.9% average F1 score. The integration of this system into existing farms, can lead to prompt interventions, curb the spread of infections and overall improvement in health management of broilers. This also research highlights the potential of computer vision in modifying poultry health monitoring practices, contributing to more sustainable and efficient poultry farming.