Crop Type Classification Using Satellite Imagery

We can detect crop types historically back to 2016 as well as in the growth season, approx. 31-37 on Zadoks scale depending on the region and crop-type. In a Norwegian context we start run the model in the middle of July and achieve our best results in the middle of August, approximately 1.5 months prior to harvesting.

Global Coverage

We’re always working hard to bring new regions like France, Spain, Brazil, US, Australia, Germany and 30+ more.

Predictable Price

We've prepared a convenient system of discounts for increasing the volume of use. More requests - less price.

Full Documentation

We provide full access to all our documentation and simple API / Add-ons endpoints for easy integration.

How do we classify all crops?

Crop Classification Model built on the baseline of crop specific yield-data. The alpha version we released in 2019 achieved a 83-87% accuracy based on single-pixel algorithm (Sentinel-1), while our latest model released we reached an accuracy of 92% based on object-based Sentinel-2 algorithm.

Different Crops

Not only major crops like barley, rye, oats and wheat - the more diversity in crop specific data will increase overall accuracy.

Crop Training Data

The Model is built on validated and reliable ground truth data from over 100,000 field level crop seeded data over a 5-year period.

High Accuracy

Our latest model reached 92% due to the increased accuracy of field delineation model and fusion between S2 and S1 data.

Project Based

We work with Crop Classification in the form of projects which fall under specificity levels, with increasing levels of difficulty.

Empowering Innovative Agriculture Companies

Crop Detection Data Delivery
Crop Detection Data Delivery
Crop Detection Data Delivery

Currently our crop detection is produced in a project-basis, so it will be pre-processed for the area of interest and then made available through API queries.

Some examples of projects:

  • Crop Classification Model in Western Australia for 2020 season based on Wheat, Barley, Oats and Canola.

  • Large-scale Crop Classification Model in India (grapes, onion and sugar cane).

  • Small-scale Crop Classification Model in Thailand on sugar cane and rice paddies.

  • Large scale Crop Classification Model on "Corn" in Myanmar.

Currently our crop detection is produced in a project-basis, so it will be pre-processed for the area of interest and then made available through API queries.

Some examples of projects:

  • Crop Classification Model in Western Australia for 2020 season based on Wheat, Barley, Oats and Canola.

  • Large-scale Crop Classification Model in India (grapes, onion and sugar cane).

  • Small-scale Crop Classification Model in Thailand on sugar cane and rice paddies.

  • Large scale Crop Classification Model on "Corn" in Myanmar.

Currently our crop detection is produced in a project-basis, so it will be pre-processed for the area of interest and then made available through API queries.

Some examples of projects:

  • Crop Classification Model in Western Australia for 2020 season based on Wheat, Barley, Oats and Canola.

  • Large-scale Crop Classification Model in India (grapes, onion and sugar cane).

  • Small-scale Crop Classification Model in Thailand on sugar cane and rice paddies.

  • Large scale Crop Classification Model on "Corn" in Myanmar.

We provide various API endpoints:

  • Data availability for a given bbox;

  • Automatically delineated low resolution field boundaries for a given bbox or as MVT vector tile (for selecting purpose);

  • Automatically delineated high resolution field boundaries with pre-processed crop-detection as metadata for a given bbox, point or field id (taken from low res boundaries).

  • Data availability for a given bbox;

  • Automatically delineated low resolution field boundaries for a given bbox or as MVT vector tile (for selecting purpose);

  • Automatically delineated high resolution field boundaries with pre-processed crop-detection as metadata for a given bbox, point or field id (taken from low res boundaries).

  • Data availability for a given bbox;

  • Automatically delineated low resolution field boundaries for a given bbox or as MVT vector tile (for selecting purpose);

  • Automatically delineated high resolution field boundaries with pre-processed crop-detection as metadata for a given bbox, point or field id (taken from low res boundaries).

Common Questions

What determines the accuracy of our crop classification model?

This is dependent on the amount of clear imagery available over a time-series, in this instance from 2015 (or 2016 if no data available in 2015) till 2021. Additionally, it will also depend on the amount of detailed ground truth training data available.

How long would it take to classify another crop?

This depends on the crop type and training data available (along with temporal data request) but if there is sufficient data available this would include 1-2 months (from the time/date of ground truth/training data and imagery is ready).

How many crops can we detect?

Good question! Yes, prior to the release of productivity zones for a given region our team will create region-specific training data so that our field delineation model produces boundaries with a high level of precision (IoU of 0.95), which will then be used in the zoning process.

How are your zones different from other options?

DigiFarm can run cloud-removal when there is no to little cloud-free Sentinel-2 imagery, however, if there is consistent periods of no clear imagery (Monsoon or similar weather patterns) the cloud-removal will not be effective, hence we would leverage Sentinel-1 SAR data. Typically, we do cloud-removal on imagery with less than 40% cloud cover.

Technical Partners That Believe In Us

How our pricing works

Our pricing is based on the data layer and the land area queried through the API. We provide different packages and options for your business. Check it now.

How our pricing works

Our pricing is based on the data layer and the land area queried through the API. We provide different packages and options for your business. Check it now.

How our pricing works

Our pricing is based on the data layer and the land area queried through the API. We provide different packages and options for your business. Check it now.