2024 · Hardware / Robotics
GreenGuardian Autonomous Weed Detection Robot
A weed-spotting agricultural rover built with a Raspberry Pi, YOLOv5, and a 3D-printed spot sprayer.
Canada-Wide Science Fair Bronze Medal · 2024
The problem
Conventional agricultural sprayers broadcast chemical herbicide across entire fields because they have no way to distinguish weeds from crops. Over 90% of what is sprayed lands on soil and healthy crops, contaminating groundwater and accelerating herbicide resistance.
The constraint
Commercial precision-spraying rigs cost hundreds of thousands of dollars. The goal was to build a working prototype on a tight budget using a single camera, a Raspberry Pi, and 3D-printed parts—while handling outdoor sunlight glare and rough soil.
What I built
Built three chassis iterations, ending with dual independent high-torque rear motors for 360-degree skid steering. Implemented adaptive histogram equalization (CLAHE) and HSV thresholding to cancel lighting shifts, paired with an 8-bit quantized YOLOv5 plant classifier running in <150ms on the Pi CPU. Connected detection to a 12V solenoid nozzle that pulses herbicide for 0.15s only over detected weeds.
The outcome
94% weed classification precision, ~90% herbicide volume reduction, and Bronze Medal at the 2024 Canada-Wide Science Fair (CWSF) in Ottawa.