Abstract:
Agriculture remains the backbone of Rwanda’s economy, yet it faces significant challenges due to inefficient traditional practices, delayed disease detection, and excessive use of inputs like water and pesticides. This thesis presents the design, fabrication, and intelligent integration of an autonomous agricultural drone system capable of pesticide spraying and maize disease detection. The system incorporates a custom-built drone platform calibrated for precision spraying and equipped with a lightweight deep learning model for real-time disease identification. Leveraging the capabilities of YOLOv8 and Google Colab, the model was trained on a curated dataset of maize leaves to classify three prevalent maize diseases: Common Rust, Gray Leaf Spot, and Blight, alongside healthy maize conditions for robust comparison. This research presents a comprehensive analysis of current stateof-the-art agricultural drone systems for disease detection and precision spraying, revealing critical gaps in real-time processing, lightweight AI deployment, and integrated actuation mechanisms. The systematic review of 2020-2024 literature demonstrates that while academic models achieve exceptional performance of 99.15% accuracy using hybrid CNN-vision transformer architecture, practical field deployment faces substantial challenges. YOLOv8 variants emergy as the optimal balance between accuracy 99.04% mAP and deployment feasibility. The model achieved a classification accuracy of 98%, outperforming traditional machine learning approaches such as Support Vector Machines, which averaged 92% accuracy in the reviewed literature. Calibration and simulation processes were conducted to validate both the spraying mechanism and model reliability under field-like conditions. Technical specifications reveal that agricultural drones with 1kg payload require 40A ESC systems with 3S LiPo compatibility with 13-35 minute flight times. This work demonstrates the potential of lightweight, AI-enabled UAVs to improve precision agriculture by enhancing early disease detection and targeted pesticide application, contributing to increased crop yield and sustainable farming practices.
