Mobile-Friendly-Based Cocoa Leaves Disease Detection Deep Learning Model Using YOLOv11Pages 106-114
Abstract:
Uncovering knowledge about rural areas is a significant challenge for the effective use of modern technological diagnostic tools, such as disease detection and diagnostic systems. It is essential to understand the extent of cocoa diseases to ensure the stability of farm productivity across Africa, especially in West Africa. This paper presents a mobile-friendly machine learning model to detect and classify anthracnose, black pod disease (BPD), cocoa swollen shoot disease (CSSD), vascular streak disease (VSD), and healthy leaflets. The proposed solution addresses challenges of accessibility to cocoa disease detection and classification machine learning models to assist cocoa farmers in improving productivity. A lightweight YOLOv11 model was developed, compatible with edge computing devices, which is easily accessible to cocoa farmers across Africa to assist farmers in detecting various known cocoa diseases affecting crop yield. The results obtained reveal a remarkable performance with a mAP@50 greater than 98% for the majority of classes and an average accuracy and recall exceeding 95%. The model has been successfully converted to TensorFlow Lite for easy deployment on mobile devices. This approach allows for automated, rapid, and accessible diagnosis for farmers, helping to improve crop health and food security.
Keywords: Cocoa disease detection, YOLOv11, Mobile-friendly deep learning model, Precision agriculture, Artificial intelligence
|
PDF 