Image Processing

Create cloud-free image composites

Create cloud-free image composites

Our software takes daily, high resolution PlanetScope images and combines them into cloud-free images for each season in a year.

High Resolution images

High Resolution images

Captured every day

Captured
every day

Cloud-free image composites

Cloud-free image composites

Imagery of all seasons

Imagery of all seasons

High quality imagery

Improves the accuracy of our mapping algorithm.

High quality imagery

Satellite captured images

plain

Labelling Platform

Labelling team
Cropland Labelling

Cropland Labelling

Each training/validation site is mapped by several trained labellers, who trace field boundaries visible in the imagery

Reducing Label Error

Reducing
Label Error

We assess labeller accuracy using a multi-dimensional scoring algorithm, allowing us to select the best label for each site, or to combine several individuals’ labels into a single “consensus” label

More Accurate Training Data

More Accurate Training Data

Machine Learning

icon
a
System identifies least certain areas ()
b
Labelling team updates labels
c
Algorithm refined and map updated
  • 1st iteration

    1st iteration

    With every iteration, the uncertain sites are identified and labelled, improving model performance

  • 2nd iteration

    2nd iteration

    With every iteration, the uncertain sites are identified and labelled, improving model performance

  • 3rd iteration

    3rd iteration

    With every iteration, the uncertain sites are identified and labelled, improving model performance

  • 4th iteration

    4th iteration

    With every iteration, the uncertain sites are identified and labelled, improving model performance

Cropland Probabilities

Cropland Probabilities

Field Boundary Delineation

Cropland Probability

Cropland Probability

Field Boundry
Crop Field boundaries

Crop Field boundaries