Abstract
Controlling weeds is essential for farmers to protect resources and maximize crop yield. Between crops, weeds are typically controlled by applying herbicides or tillage to the entire field. However, these control methods are expensive and can pose environmental risks. Robotic weeding systems are a good solution to minimize environmental impact and save money on herbicides, but they are expensive (>$100,000). The Weed Warden is a low-cost (<$200) plant detection sensor that can be mounted on rovers or tractors. The Weed Warden uses an open source multispectral sensor to detect live vegetation and sends a logic signal that could trigger a weed removal system such as a sprayer or mechanical tillage when vegetation is detected. We evaluate the Normalized Difference Vegetation Index (NDVI), Enhanced Normalized Difference Vegetation Index (ENDVI), and Enhanced Vegetation Index (EVI), for producing a value that, combined with a calibrated threshold, will indicate if there is plant life under the sensor. The Weed Warden system using ENDVI is most consistent at detection, with the ability to discriminate 7.6x7.6 cm vegetation samples from bare soil at sensor heights of 30 and 41 cm from the ground. The Weed Warden is a proof-of-concept component of an alternative system to robotic weeders of fallow fields that could help reduce costs, improve environmental outcomes in agricultural settings, and advance research into fallow field management practices.