The usage of herbicides for weed control is waning in the last years due to their environmental impact, uncertain impact on human health and the growing concern about spread of resistant species as well as due to growing demand for organic produce. This creates a large market for cost-efficient weed control methods and was one of the reasons a computer vision based precise system that can make old concept of inter-row commercially viable is so valuable.
Inter-row weeding is the practice of controlling the weed growing between the rows of crop plants and can be dated as far back as the 18th century but with the introduction of herbicides fell out of favor and today is essentially non-existent in most of the world.
The main problem this technique faces is that in attempting to weed out as many as the weeds we might hoe the crop plants themselves and there is a real value in being able to get as close as possible to the plants without actually weeding them. Naturally precision in itself is not enough as we also want to cover as much land as possible in the shortest possible time and so a tradeoff between precision and speed is also taking its place in the problem.
The solution presented here relies on the amazing leaps computer vision tools have gone through in the past decade by opting to use cheap cameras. This approach allows for a flexible solution in respect to types of crops and growth staged as well as produces a cheaper product to deploy by moving the entire processing responsibility to the software side. Learn more hear about computer vision for weed management.
When dealing with the processing of the images they have encountered three major challenges – naturally varying lighting, incomplete crop growth (i.e. inconsistency in the rows themselves) and weeds occluding the crop plants.
To deal with this problems they used prior knowledge of the row spacing which can be produced either manually or by keeping GPS records of previous work done in the field.
To enhance their precision they also incorporated tracking of the rows between the images which allowed them to have a consistent result throughout the operations.
Finally instead of using a single camera they used an array of cameras which allowed them to capture several rows at once, giving them the ability to share the knowledge and data and eliminating the effect of local changes due to crop or weed density variations.