Imaging stale seedbeds for weed mapping, monitoring and early season control

Weed detection systems are an active area of research, and several systems are being brought to market to enable targeted spot spraying systems, that can dramatically reduce the amount of herbicide used (and so reduce the environmental and financial costs). However, these systems typically work on weeds after the crop emerges. This is an important short coming for spot spraying because many key herbicides are applied before the crop emerges (called pre-emergent herbicide). In addition, a key use cases of weed detection is to map weed populations over time to assess management strategies and emerging resistance. By the time the crop and weeds have emerged the majority of weed control actions have already been carried out. Thus, a weed detection system that works in crop can only be used to monitor post-control weed populations, greatly reducing the ability to monitor management effectiveness.

This PhD will develop a machine-vision pipeline to detect weeds in the pre-crop fallow period, known as a the stale-seedbed. Successfully detecting and mapping weeds in the stale-seedbed will provide a direct measure of the underlying seed bank (composition, distribution and abundance). The seedbank is the most important measure of long-term weed populations because the seed bank is the only means of long-term persistence for almost all arable weeds.