Land Lines Image Segmentation

The Land Lines Image Segmentation (LLIS) process block generates GeoJSON polygons that delineate major land cover classes, vegetation management zones, and natural land surface features from sentinel-2 satellite imagery.


The Land Lines Image Segmentation processing block generates GeoJSON polygons that follow major land surface features from Sentinel-2 satellite imagery, including water bodies, rivers, vegetation cover and types (forest, shrubs, grassland, crops, etc), and other visually distinct classes. This linework can be used as an input for feature classification workflows or to vectorize imagery into useful map products.

The output includes both large and small land line polygons and the sentinel-2 RGB image of the user scene as well for linework reference.

Technical Description

Land Lines Image Segmentation requires multispectral imagery to develop linework for delineating land, water, and vegetation boundaries. The outputs can be used in object-based image analysis workflows which are known to provide better land cover and image feature classification than pixel-based analysis for vegetation zonal management, including forestry inventory stand applications.

Large land lines are useful for gross land cover delineation and vegetation disturbances (forest cutblocks or crop loss events), while the small land lines are useful for finer feature delineation like smaller streams, right-of-ways, and vegetation corridors. The average polygon size is 10 ha for large segment output and 1 ha for small segment output results. Features smaller than 10 pixels will not be delineated. Maximum area per run is equal to full sentinel scene (~10,000 km2).

Known limitations: clouds, shadows, and snow will influence the results. Since the ‘max_cloud_cover’ filter for the input imagery applies to the full sentinel-2 tile and not just the user specified bounding box, it is suggested to set this filter to 0 or less than 5% for best results.

Algorithm Training Data Details

This process was developed over several years of research automating image segmentation workflows for high accuracy forest stand and block delineation using lidar and aerial imagery to create uniform stand boundaries for precision forestry management (see for more info). This adaptation of this algorithm was trained on Sentinel-2 imagery from several dozen cloud-free scenes ranging from 10 km2 to 10,000 km2 from natural landscapes all over the world including Canada, USA, Brazil, Chile, Europe, China, and India.

Terms & Conditions

View the End User License Agreement conditions.
Requires additional partner approval.

Ready to get started?

Get in touch or become a partner.

Contact salesBecome a partner