AI applications for wine and vine

BigDataGrapes aims to help European companies in the wine and natural cosmetics sectors to become more competitive in international markets. In particular, it aims to help companies in the wine value chain to benefit from Big Data tools, by providing decision support resulting from the real-time and transversal analysis of important and diverse data sources. For example, as part of the BigDataGrape project, works based on deep learning techniques have been carried out to develop a tool capable, from images, of automatically counting the number of leaves per vinestock. This tool can be used in several application frameworks, such as to feed agronomic models, particularly 3D representation models, or for precision viticulture applications, particularly to limit the use of inputs. This work is being carried out by the CNR in Italy and the LEPSE and MISTEA units of INRAE in France.

As part of the project, the CNR teams in Italy have also developed machine learning methods for predictive data analytics on wine data, collected from online social networks of wine passionate users (further analyzed in deliverable 4.3 of BigDataGrapes project). The data available consisted of a general description of the wine, user-generated data (notes and comments) as well as user profile information. A set of algorithms was deployed to assess the potential market penetration of a given wine in a new country. They estimated this penetration capability by learning a model, from user-generated content, to be able to predict wine ratings in a target country from wine characteristics (grape variety, origin, aromas, etc). Within the context of the project, the University of KULeuven is collaborating with INRAE - MISTEA to develop visual analytics tools to explore large datasets and to present meaning emerging from data. A tool is being developed to visualize variables related to viticulture and winemaking for different wines over several years. The user will be able to select targeted aromas or specific characteristics (grape variety, years of interest, etc.) and compare wines produced using selected variables from vine to wine. To achieve this goal, the preliminary work was semantic data linking in order to connect them in an intelligent way, to guarantee a vine-grape-wine continuum, based on experimental data from Pech Rouge (INRAE unit). This tool will enable us to answer several questions such as the influence of climatic effects or the winemaking process on the aromatic composition of wine.

As part of a collaboration between MISTEA and NYSEOS, a study was conducted to better understand the environmental effects on the aromatic profiles of wines. Measurements of aromatic compounds on a learning set of several dozen different wines were used. The wines studied came from different sites and different vintages and 3 grape varieties were used. For a given grape variety, it is always the same type of practice and the same type of winemaking that was carried out regardless of the site or the vintage. The large number of aromatic compounds in the wine as well as the high heterogeneity of the data makes it difficult to characterize the aromatic profiles and interpret them using classical methods. To conduct this study, the data were structured in the form of a graph, and the idea followed was to look for a minimum subset (a key from the computer point of view) of compounds capable of discriminating different wines. This can be interpreted as the aromatic imprint of a wine. This work led to the development of a tool capable of learning how to distinguish wines and trying to mimic the behavior of an expert. The tool can learn the aromatic fingerprint of hundreds, even thousands of different wines. It can then establish links between subsets of aromatic compounds and different datasets describing environmental conditions.

Many decision support tools, based on artificial intelligence methods and relying on Big Data tools, are currently under development. For example, we can cite a tool for the early detection of disease (flavescence dorée) using AI techniques (Chouette company) or a virtual wine shop service (Matcha company). The major issue for the development of this type of solution remains data. The wine-growing environment is in the midst of a technological transition and many players in the sector need to benefit from new advances to meet challenges such as adaptation to climate change or the reduction of inputs. To achieve this goal, it is crucial to be able to gather data and make them accessible to wider communities in a not complex or redundant way, but this is a real issue and a fundamental step for the development of all current and future intelligent tools.

(The photo is provided by Dan Meyers from Unsplash)

Image: