<![CDATA[Purpose – This study implements the YOLOv9 architecture to automatically classify the health condition of ganyong leaves (Canna edulis Kerr.) as an early-detection tool for leaf spot disease. The study addresses the limitations of subjective manual identification and supports farmers in Bumipajo Village, Bima Regency, in reducing potential crop failure. Methods – A primary field dataset consisting of 1,383 image objects was collected and divided into training, validation, and testing sets using a 70:20:10 ratio. YOLOv9 was implemented by integrating Programmable Gradient Information (PGI) and the Generalized Efficient Layer Aggregation Network (GELAN). Model training was conducted in Google Colab using GPU acceleration, a batch size of 4, and 50 epochs. Findings – Evaluation on independent test data showed strong detection performance, with mAP@50 of 99%, Precision of 99%, Recall of 100%, and an average inference speed of 58.4 ms per image. These results indicate that YOLOv9 can effectively preserve disease-related morphological features in visually complex biological objects. Research implications – The findings are limited to the environmental conditions of the data collection site and one disease type. The reported time efficiency also depends on GPU-based hardware and requires further validation on mobile devices. Originality – This study contributes a field-based primary dataset of ganyong leaves and validates YOLOv9 for a local agricultural commodity that remains underexplored.]]>
Copyrights © 2026
