An object detection solution for early detection of taro leaf blight disease in the West African sub-region
Chidiebere Nwaneto, Chika Yinka-Banjo, Ogban-Asuquo Ugot
Abstract
Taro Leaf Blight (TLB) poses a significant threat to food security and economic stability in West Africa, where taro is a staple crop. This research presents an object detection system utilizing the YOLOv8 deep learning model to detect TLB early in taro plants. The methodology involved developing a unique dataset comprising images of taro leaves at various stages of infection, collected from farms in Nigeria and Ghana. Fine-tuning the YOLOv8 model with this dataset resulted in a notable improvement, achieving an 85.7% mean Average Precision (mAP) across all classes—a significant enhancement over existing generic plant disease detection models, which typically achieve mAP values of around 70-75% on similar datasets. This 15-20% improvement enables more accurate early detection, crucial for timely interventions. The system was subsequently integrated into an Android application, allowing farmers real-time diagnosis and disease management access. Field tests demonstrated the application's effectiveness and user-friendly design, making it a practical tool for early disease intervention. This research highlights the potential of combining deep learning and mobile technology to address agricultural challenges and improve food security in the region.