Analysis of the performance of the detection of cassava leaf diseases using YOLO versioning
- International Conference of Science and Information Technology in Smart Administration (ICSINTESA) : 678-683
Résumé
Early detection of cassava leaf diseases is important for preserving agricultural yields in regions where this crop is an essential food source, particularly in Sub-Saharan Africa, where losses reach 20-30% due to diseases such as Cassava Mosaic Disease (CMD), Brown Streak Disease (CBSD), Green Mites (CGM), and Bacterial Blight (CBB). This study evaluates the performance of nano/tiny variants of five recent YOLO models (YOLOv8n, YOLOv9t, YOLOv10n, YOLOv11n, and YOLOv12n) for the automated detection of these diseases on an annotated dataset collected from Roboflow and the local PLANTPATHO-INFOS-BURKINA FASO platform. The results show that YOLOv12n offers the best performance, surpassing previous versions in accuracy while maintaining suitable real-time speeds on resource-constrained hardware like Tesla T4 GPUs. These improvements, combined with a qualitative analysis on a test image, highlight the potential of YOLOv12n for real-time detection applications, paving the way for practical tools for farmers in low-resource settings. This study fills a critical gap in the literature by providing the first systematic comparison of YOLOv8 to YOLOv12 on a cassava-specific dataset from West Africa, addressing the lack of evaluations on local, real-world agricultural data.
Mots-clés
cassava leaf disease, plant disease detection, YOLO, object detection, detection performance evaluation