<![CDATA[Porang (Amorphophallus muelleri) is a high-value agricultural commodity whose productivity can be significantly reduced by leaf diseases. Conventional visual inspection is time-consuming and subjective, highlighting the need for an automated detection system that is both accurate and efficient. Previous studies on porang leaf diseases have primarily focused on image classification or object detection methods with limited capability to generalize under varying visual conditions. Therefore, this study aims to evaluate the performance of the YOLOv8n model integrated with data augmentation for detecting and classifying five conditions of porang leaves: early blight, late blight, konjac mosaic, insect attacks, and healthy leaves. The proposed approach combines YOLOv8n with on-the-fly data augmentation during training and Stratified 5-Fold Cross-Validation to improve the model's generalization capability. A total of 1,065 leaf images were used for model training and evaluation. Data augmentation techniques, including Mosaic, HSV color adjustment, and geometric transformations, were applied during training. Experimental results show that the proposed model achieved a mean Average Precision (mAP50) of 80.53%, a Precision of 83.11%, and a Recall of 70.66%, outperforming the baseline model without augmentation, which obtained an mAP50 of only 63.15%. Although the early blight and healthy classes achieved excellent detection performance (mAP50 > 90%), the relatively lower Recall was mainly caused by background bias in the insect class, resulting in several objects being missed during detection. Overall, the integration of YOLOv8n with data augmentation demonstrates its potential to improve the generalization capability of porang leaf disease detection systems under diverse visual conditions.]]>
Copyrights © 2026
