PoultryNet: A Lightweight CNN Student Model for Chicken Disease Classification via Knowledge Distillation from Vision Transformer

Poultry diseases seriously endanger the poultry industry, leading to economic losses, as well as undermining food security. While there are various precise heavyweight disease detection models, there are few fast, lightweight, as well as deployable, solutions that are appropriate for real-world farm deployments. This research bridges this gap by introducing PoultryNet, a tailored lightweight Convolutional Neural Network (CNN) trained as a student network through Knowledge Distillation (KD) from a Vision Transformer (ViT-B/16) teacher. While the teacher attains 98.83 % accuracy on a Kaggle chicken disease dataset containing four disease classes, PoultryNet achieves a reasonable test accuracy of 95.93%, presenting a fast and resource-optimized alternative. Extensive experiments confirm that PoultryNet not only surpasses the best classification performance on all four classes…