Development of a machine learning-based disease identification algorithm for Robusta coffee

Abstract

Robusta coffee is the main cash crop grown and a significant contributor to Uganda’s economy, today coffee faces persistent challenges caused by diseases affecting coffee plants. This project aims to build an edge device that farmers will use to detect coffee leave rust. During the course of this project, a local dataset of 1000 images was created and made publicly available. A machine learning model was built using YOLO v8, and deployed on a Raspberry Pi 4 to which a pi camera was attached. Due to the limited data collected, and limitations in computational resources available during the execution of the project, our model accuracy was low at 62%. After making some additional consultations, consultants were optimistic on the project results, however, the model accuracy was not able to be improved due to budgetary and time constraints. Although the accuracy was low, the results demonstrated a starting point and act as validation that machine learning models that can detect small artifacts for plant disease detection are a possibility. And with more data and most importantly with the availability of more powerful computational resources available to researches and hobbyist, model accuracies in plant disease detection is only bound to increase.