The Blog


A couple of years ago remote sensing made its way into agriculture and it has since been powering a wide range of applications, supporting farmers to monitor their fields, helping regulators to identify fields and crop species and finally traders to follow trends and potential yield.

In recent years actionable information that used to mainly rely on satellite imagery has been pushed aside by the recent boom of drones, now widely used by farmers around the world in their constant search to improve productivity.

What about vineyards?

Wine producers also rely on actionable information to make critical decisions that will help maximise the quality and yield of their production, without talking about the need to sustainably manage their vineyard. With factors such as increasing international competition, price volatility and climate change affecting both wine production and the taste, it is extremely difficult and time consuming to gather detailed information on thousands of vine crops every day.

Historically, due to the row structure and small size (less than a meter width) vines were not good candidates for monitoring applications based on Earth observation technologies. The increasing availability and affordability of Ultra and Very High Resolution images drastically changed the game.

(left) 3cm resolution drone image, (middle) 30cm VHR satellite, (right) 3.5m HR satellite like Planetscope

Nowadays, with drone imagery providing centimetre resolution and VHR satellite imagery reaching 30cm resolution, vineyards can be mapped accurately and information on vines can be extracted using the appropriate image processing pipelines. Fact-based decisions to harvest, treat or replant are therefore becoming available to every vineyard manager.

Picterra’s solution for smart vineyard monitoring

At Picterra we intend to democratise the access to machine learning tools and remote sensing data to help improve and speed up geospatial analysis. We are integrating in our platform ( the possibility to extract some of the critical information about a vineyard. To begin, the user will need to provide images with a NIR, Red and Green bands and optionally a Blue band as well as a vector file (GeoJSON, KML, …) containing the parcels of interest.

Mapping vine rows

The first step is to detect the general orientation for each parcel. This robustifies the detection of the rows and vines afterward. If the orientation of a parcel is wrongly detected it is possible to adjust it in a few clicks.

Based on the orientation the detection of the vine rows at each parcel is performed. As before, if a row is missing or wrongly detected this can easily be adjusted by interacting with the map.

False colour image above vineyard parcel and the detected rows overlayed (clear blue).

Detecting missing vines

From the vine rows, the algorithm will detect vines and missing vines along each row. As the foliage of vines can be pretty dense and does not always allows to distinguish vines from each other, a default inter-vine distance is exploited. Our machine learning algorithm is semi-supervised, exploiting pre-trained examples along with examples from the inter-row region and over the rows for the parcel. Thanks to this it can adapt todifferent type of parcels.

The detected missing vines (red cross) and present vines (green circle) overlayed on the false color image.

Vine vigour

The interesting aspect of having vines locations, besides spotting missing plants, is to extract vegetation information over each plant. Vegetation indices can be extracted to monitor the relative variations of vigour in a parcel and between parcels. Statistics about vine vigour and missing plants can easily be derived from this information.

Vine plants overlayed on a vegetation index (NDVI) colormap

One of the best way to visualise vines vigour is to interpolate the information over the parcel as shown in the image below. This cleans the information and removes noise from the inter-row region.

Vine viguour (NDVI vegetation index) interpolated over the parcel. Only vine information is shown, removing all perturbations from the inter-row region.

From information to actions

All this information can support decision and actions: from reporting to actual field work. The statistics on the occupancy (number of vines over the number of present and missing vines) or the mean viguour of a parcel can be derived in order to get an overview of a domain and focus the work on certain parcels.

At the time of harvest the vine vigour can help to efficiently schedule the work. At the time of replanting vines, this allows to estimate the number of vines needed. At the time of treating for pests or fertilization, this helps you to estimate the amounts required for each parts of your exploitation.

How to access it?

Interested in having this type of analytics run on your imagery? Sign up to and request the vineyard plugin or contact us at

%d bloggers like this: