← Marketplace

Land cover classifier for Pléiades/SPOT


Coming back soon: currently not available. Classifies imagery into discrete land cover classes.

Technical information

Land cover classifier for Pléiades/SPOT is an analytics algorithm that utilizes a deep learning network for land cover classification.

Technical Information

Land cover classification or segmentation is the process of assigning each of the input imagery pixels a discrete land cover class (e.g. water, forest, urban, desert etc.).

The land cover classifier is trained with Pléiades/SPOT imagery over multiple locations and can be applied globally. The algorithm accepts 4 or 5 as number of classes to be inferred. In case of 4 classes the model treats Barren Land and Urban as a single class. Currently, the classification output consists of the following classes:

  1. Water
  2. High vegetation (including trees)
  3. Low vegetation (including bushes and grass)
  4. Barren Land
  5. Urban (including roads and buildings)

Based on an independent test dataset we reached an Accuracy of 0.75 and a Jaccard of 0.22 when trained with 4 classes. When trained with 5 classes using the same test dataset, we achieved an Accuracy of 0.64 and a Jaccard of 0.51. The algorithm can be used globally.

For more information about the development of this land cover segmentation classifier, see the published blog post which describes our approach. The used model architecture is based on the work of Robinson2019. The basis of the model training code is publicly available in this repository.

Terms & conditions

View the End User License Agreement conditions

Ready to get started?

Explore UP42 now.

Sign in or create an account