A new project will create a system in which a camera and sprayer work together to automatically recognise weeds in the field and only spray when necessary.
2017.08.07 |
Weeds can be avoided and herbicide use curbed significantly with the aid of new technology. Photo: Janne Hansen
Artificial intelligence will become more common in Danish agriculture in the years to come – well assisted and supported by researchers from Aarhus University. They are collaborating with a range of commercial companies on developing new technology for precision weed control that can reduce herbicide use by up to 75 percent.
This will take place in a new project to which Innovation Fund Denmark has granted 19 m DKK and which is led by the technology company Datalogisk. The overarching aim of the project is to achieve significant savings on herbicides to the benefit of the environment, agriculture and society.
Basically, the system works by having a sprayer carrying a camera mounted on the back of an ATV. The camera can recognise the individual weed plant when the sprayer passes over it and the sprayer is only activated when it is necessary. In this way, selected weeds are controlled in a very targeted manner.
The RoboWeedMaPS project stands on the shoulders of previous research from Aarhus University regarding optimisation of weed recognition. The project partners see great potential in taking the research several steps further via computer vision, artificial intelligence and big data.
Parts working together for a whole
The project will develop several components that can be used independently or integrated. All told, the project comprises:
In unison, the camera and the other products will constitute a total, easily handled and simple product for farmers that can have a big effect on their management and finances. The generic qualities of all the products make them well suited for export via already established networks in Europe.
Extensive weed memory
The product will consist of both a software and hardware component. The core of the software component will be to develop a system that can automatically recognise the weeds and weed species that are detected in the field. This is where deep learning will be put into action. This means that a neural network (artificial intelligence) will be presented with and store large amounts of data – so-called big data. In this way, the network will learn to understand what to look for in the data and to recognise or find new connections based on what it has already seen.
In this case the big data will include thousands of images of various types of weeds. The camera system must be able to recognise all kinds of weeds in all kinds of growth stages. The images, that can be gathered directly from the sprayer in the field, by drone or by mobile telephone, can automatically be transformed into weed maps and targeted spraying maps adapted to the farmer’s existing technology. The farmer’s investment will thereby be kept to a minimum compared to the savings that will be seen on herbicides.
Facts about RoboWeedMaPS - Automated Weed detection, Mapping and Variable Precision Control of Weeds:
For more information please contact: Senior Researcher Solvejg Mathiassen, Department of Agroecology, telephone: +45 8715 8197, e-mail: sma@agro.au.dk
