Towards Practical Precision Agriculture: Lemon Leaf Disease Detection Using YOLO11n and a Real-World Dataset

Abstract

Lemon production plays an important economic role in Vietnam; however, leaf diseases severely reduce plant vigor, yield, and fruit quality, posing significant challenges to precision agriculture. De spite recent advances, automated lemon leaf disease detection re mains underexplored due to the lack of high-quality, lemon-specific datasets and efficient, lightweight models suitable for real-world de ployment. To address this gap, this study introduces LemonDisease 5i, a new real-world dataset comprising five classes, including four common lemon leaf diseases and one healthy class, carefully col lected and lesion-level annotated by agricultural experts. Based on this dataset, a lightweight YOLO11n-based detection framework is developed to enable robust multi-scale feature extraction un der complex field conditions. Experimental results demonstrate that YOLO11n achieves superior performance, with a precision of 89.9%, recall of 88.9%, and mAP@0.5 of 93.2%, outperforming YOLOv8n, YOLOv9t, and YOLOv10n. These findings confirm both the effectiveness of the proposed model and the suitability of the LemonDisease-5i dataset as a reliable benchmark for real-time pre cision agriculture applications

Keywords