<![CDATA[Banana leaf disease (Musa spp.) poses a critical threat to tropical agricultural productivity in Indonesia, with harvest loss estimates reaching 30–50% during rainy seasons due to undetected pathogen infections. Limited manual diagnostic capacity among farmers produces disease misidentification and delayed control interventions, particularly at early infection stages when inter-class visual symptom similarity remains high. This study proposes an automated banana leaf disease detection system leveraging deep learning through the You Only Look Once version 5 (YOLOv5) architecture as a digital image-based diagnostic solution. The dataset comprises 9,684 images across eight banana leaf disease classes, curated via Roboflow with 87% training (8,460 images), 8% validation (820 images), and 4% testing (404 images) splits. Preprocessing includes 2×2 tiling, auto-orientation, and 512×512 pixel stretch resizing. Data augmentation applies horizontal and vertical flip alongside 0–25% zoom crop, generating three outputs per training image. Model performance evaluation employs precision, recall, F1-score, and mean Average Precision (mAP@0.5) metrics. Results demonstrate YOLOv5 capability in accurately detecting and classifying banana leaf diseases under Indonesian tropical field imaging conditions.]]>
Copyrights © 2026
