Assessment of green recognition for automatic generation of variable-rate prescription weed maps

Variable-rate spray-application technology is now commonplace in precision agriculture (PA). However, this technology cannot automatically produce detailed herbicide-application maps. Recently, technology around unmanned aerial vehicles (UAV) and image sensors has advanced to where low-cost consumer-grade UAV systems can provide relatively high-quality imagery. This has led to the many farms using these UAV systems to monitor different aspects of farming activities including weed populations. Using machine vision and consumer-grade UAV technology, the University of Southern Queensland (USQ) is currently developing software to automatically generate prescription spray maps and provide an inexpensive alternative to the current ‘see and spray’ herbicide-application systems. We describe a preliminary test of weed-detection capabilities using machine vision and low-cost UAVs at different altitudes. Results show that higher green-sensitivity values improved weed recognition for all heights. In addition, colour loss at higher altitudes has a greater impact on weed-recognition performances than spatial resolution alone when using green-from-brown differentiation. These findings will help further development of the site-specific weed management package currently being developed by USQ. Key words Precision agriculture, machine vision, sensing, unmanned aerial vehicles, weeds