The quest for accuracy in digital agriculture
While digital technology has been gaining ground in the agricultural sector for several years, existing intelligent agricultural technology, which must constantly decide on the action to implement while taking into account existing data, horticultural knowledge, laws, regulations and the specificities of each farm, is currently reaching its limits. This is due to the fact that the data such systems rely on is often incomplete or subject to error, or that sometimes very different contexts prevent the systems from using the data correctly.
For example, distinguishing between undesirable weeds and those which should be spared thanks to the various agro-ecological services they provide represents a task which exceeds the abilities of current technology. Furthermore, autonomous systems are unable to provide a reliable identification of plants based on sensor data, nor can they decide which plant is a harmless or even beneficial weed for a particular crop in a particular condition.
Additional expertise for AI adapted to a complex field
The R4Agri (Reasoning on Agricultural Data) project was inspired by this realisation, and is the fruit of joint studies between INRIA and the DFKI in artificial intelligence. “The idea was to see if we could combine our respective expertise to tackle new problems linked to digital agriculture”, Marie-Laure Mugnier, co-director of the project, explains. “We discovered a mutual interest in the integration of heterogeneous data thanks to a semantic layer which can interpret, integrate and analyse such data in order to provide high-level services.”
Three teams were thus involved in this new project: the INRIA BOREAL team (formerly GraphIK), specialised in knowledge representation languages and automatic reasoning techniques for the use of heterogeneous data; and two DFKI teams: PBR for robot control via AI techniques and SDS for the development of semantic technology in the use of major sources of multimedia data. These two teams boast considerable experience in the development of applications for digital agriculture.
The two DFKI teams provide solid applied expertise focused on transfer, but are equally interested in developing techniques which are sufficiently generic to be applied in various applicative contexts. They also have access to data sourced from numerous projects. We provide more fundamental expertise and see digital agriculture as a highly challenging field of application.
co-director of the project
A springboard towards new European research projects
The aim of the project is to provide a theoretical, algorithmic and software framework to integrate and interpret various existing data sources (sensor data from farm machinery, satellite and weather data, horticultural science data, etc.) in a high-level language which enables reasoning. In other words, to enable agricultural technology to interpret data more accurately and thus make the right decisions.
“The raw data from sensors is naturally quantitative, of variable granularity and subject to error, and must thus be interpreted and amalgamated to be transformed into high-level information. In practice, this is achieved by more or less ad hoc techniques; we were searching for a more generic and declarative approach, which would take into account the context of interpretation explicitly. In fact, the same piece of data can be interpreted differently according to what we already know of the context”, Marie-Laure Mugnier explains.
The project, initiated in January 2022, is still in its structural phase. In keeping with its fundamental research aspect, the results of R4Agri will consist mainly of joint publications in scientific journals. The results obtained can then serve as a springboard for other European-wide projects in the field of intelligent agriculture and farm robotics. “And of course, the societal impact of AI applications in agriculture is huge”, Marie-Laure Mugnier concludes.