Integrating Thermal and Fecal Image Analysis Using Deep Learning for Early Detection of Poultry Diseases

Abstract

Timely and accurate detection of poultry diseases is vital for minimizing economic losses and safeguarding animal welfare in the poultry industry. Conventional diagnostic methods are often invasive, time-consuming, and reliant on human observation, leading to potential delays in intervention. This study explores the application of artificial intelligence (AI) and non-invasive technologies to enhance disease diagnosis in poultry. A thermal imaging-based framework was developed to detect Avian Influenza (AI) and Newcastle Disease (ND) within 24 h of infection. Thermal data collected from both infected and healthy broilers were processed to extract 23 statistical features, which were then analyzed using Support Vector Machines (SVM) and Artificial Neural Networks (ANN). SVM outperformed ANN, achieving classification accuracies of 97.2 percent for AI and 100 percent for ND. To address uncertain predictions, the Dempster-Shafer evidence theory was employed to enhance decision reliability. Additionally, the study proposes a deep learning-based model using fecal image classification through CNN architectures such as MobileNetV2 and Xception to identify Coccidiosis, Salmonella, and ND, achieving accuracies exceeding 98 percent. The integration of thermal imaging, fecal image analysis, and AI-based classification presents a robust, non-invasive, and rapid disease detection system. This multi-modal approach demonstrates significant potential for improving poultry health monitoring and enabling early, automated interventions.