Optimizing poultry disease classification: A feature-based transfer learning approach

The poultry industry is under pressure to meet the rising global demand for poultry products, driven by demographic trends and technological advancements in Industry 4.0. However, this growth introduces significant risks, including heightened vulnerability to disease outbreaks in densely populated farms, making robust disease management systems essential. Despite advancements in deep learning for poultry disease classification, challenges remain in optimizing models for diverse environmental conditions and disease manifestations while maintaining computational efficiency. This study systematically evaluates nine pre-trained Convolutional Neural Network (CNN) architectures, namely DenseNet201, InceptionResNetV2, InceptionV3, MobileNetV2, MobileNetV3Small, NASNetMobile, ResNet152, VGG19 and Xception combined with three machine learning classifiers, namely Support Vector Machines, Logistic Regression and k-nearest Neighbor to develop an optimised framework for poultry disease classification through faecal images, balancing computational efficiency and diagnostic accuracy for practical deployment. It was demonstrated that the ResNet152-SVM achieved the highest test classification accuracy (CA) of 98.3 %. DenseNet201-LR and DenseNet201-SVM closely followed, achieving test CAs of 97.9 % and 97.7 %, respectively. Based on the findings of the paper, it is evident that the ResNet152-SVM transfer learning pipeline could further facilitate the detection and classification of poultry diseases.