AI4Agriculture

Target audience

Wine grape producers, to ensure harvest quality.

Project objective

To ensure the quality of the harvest in the Ribera del Duero wine-growing region through computer vision and other AI techniques, with three objectives:

1. Assess grape production quality;

2. Count the number of fruits (grapes);

3. Predict yield.

Project description

The pilot is trained and validated using data from Spain's Ribera del Duero wine-growing region, which is a registered designation of origin. This imposes certain restrictions that directly affect the pilot's objectives and use cases. In particular, there is a limit on the maximum production of a vineyard, traditionally set at 7,000 kilograms of grapes per hectare. This means that grapes often have to be removed from the vines to avoid exceeding the permitted production limit. Therefore, the pilot does not aim to maximise yield, but rather to ensure quality.

It was agreed with some selected vineyards that grapes would not be removed from a small area of the plots, in order to comply with the Designation of Origin regulations. This allows us to extrapolate the results from these areas to a larger vineyard.

The current methodology begins by agreeing with the vineyards on the selection of three plots. Each will represent a different production level: high, medium and low. These levels will be defined based on production results from previous years. Later, some rows will be designated as a “no-thinning” area, where grapes will not be removed from the vines before harvesting. We will use models to select the most suitable areas for data collection, using, for example, NDVI (a multispectral index for plant vigour) and other variables (i.e. soil, plant variety, planting year, etc.). During the summer of 2020, in the months following veraison (the change in fruit colour), we expect to photograph and analyse plants and fruit, probably on a weekly basis, using the SMR mobile data collection application. In parallel, drones equipped with a multispectral sensor will be flown to obtain NDVI measurements for the sampled plants. These data will subsequently be cross-referenced with the vineyard's actual production to determine how many kilograms of grapes those plants produced.

To assess fruit quality, we will use the same samples to try to identify pests or diseases that could affect final production. If permitted, we would also like to use vineyard data, such as laboratory analyses of grapes and wine, fertilisation data, soil data and plot information (plant varieties, planting year, etc.). These data will be combined with multispectral data from drone flights to correlate NDVI with the quality of the final product.

Project funded by the European Union's Horizon 2020 research and innovation programme.