Design & Fabrication of Autonomous Weed Removing Robot Using Deep Learning

Abstract

The autonomous weeding robot, designed for efficient weed management in agriculture, features a robust rectangular frame with 4-inch wheels and a dynamic rotatory weeding blade powered by separate DC motors for vertical movement and rotation. Propelled by high-performance 1000 RPM DC motors, it operates on a two-wheel drive configuration controlled by an Arduino Uno microcontroller. It has a front-mounted camera and colour sensor; the robot exhibits superior perception capabilities for crop row detection and weed classification. Leveraging the YOLOv8m deep learning framework, it autonomously identifies crop rows with an average weed detection accuracy of 92.5%. Through rigorous experimentation, the robot demonstrates promising numerical insights, achieving an average weed removal efficiency of over 85% while traversing 15 meters in approximately 7 minutes, which reduces the cost of weeding up to 60%. These results underscore the robot's potential to significantly reduce manual labour and environmental impact in agriculture while enhancing overall efficiency.