An Efficient YOLO26-Based Framework for Banana Leaf Disease Identification

Abstract

Banana cultivation is affected by several leaf diseases that can alter leaf colour, texture and visible surface patterns. Identifying these symptoms early is useful for crop monitoring and disease management. In practice, diagnosis based only on visual inspection can be time-consuming and may vary with the experience of the observer. This study presents a deep-learning approach based on YOLO26 for automatic identification of banana leaf diseases. We consider four categories: Cordana, Healthy, Pestalotiopsis, and Sigatoka. The study uses banana leaf images obtained from the BananaLSD/Kaggle-related collection and a separate set of 768 segmented leaf images prepared using Roboflow. In the available four-class evaluation, 1,600 images were assessed, with 400 images in each class. The experiment achieved 96.88% accuracy, 96.93% precision, 96.88% recall and 96.88% F1-score. The class-wise F1-scores were 0.98 for Cordana, 0.96 for Healthy, 0.96 for Pestalotiopsis and 0.98 for Sigatoka. These results indicate that the proposed approach can distinguish the selected banana leaf disease categories with high consistency. The work is intentionally limited to one research objective: accurate identification of banana leaf diseases. Detection of disease location, object localisation and segmentation are not treated as separate research objectives.