Application of artificial intelligence for pest and disease detection on jackfruit tree trunks in Vietnam

This study addresses the critical challenge of automated disease detection in jackfruit cultivation by constructing a specialized image dataset focused on trunk pathologies—a domain significantly less explored than leaf or fruit analysis. The dataset comprises 17,728 high-quality images (post-augmentation) collected from real-world orchards in Quang Nam and Da Nang, Vietnam. It features four prevalent trunk disease classes: Batocera rufomaculata (stem borer), pink disease, stem cracking gummosis, and stripe canker, captured under diverse lighting, angles, and occlusion levels. A comprehensive comparative evaluation was conducted among five state-of-the-art deep learning architectures: YOLO11m-cls, YOLO26m-cls, Vision Transformer (ViT), EfficientNet, and ResNet50. Quantitative results on an independent test set reveal that YOLO26m-cls achieves the superior performance, reaching an accuracy of approximately 97.8%, a macro F1-score of 97.5%, and an exceptional inference speed of 6–10 ms per image. While ViT and YOLO11m-cls delivered competitive accuracies of 96.5% and 96.2% respectively, YOLO26m-cls demonstrated the most favorable balance between computational efficiency and classification robustness. Practical validation through a web-based deployment platform further confirms the model's capability for real-time diagnostic support. This research provides a foundational digital resource for smart agriculture in Vietnam, empowering farmers with a low-cost, AI-driven tool to mitigate economic losses and promote sustainable orchard management.