Detecting forest layers with AI

AI Engineering


Marcus Gullstrand
LinkedInlänk till annan webbplats, öppnas i nytt fönster

Mail
GitHublänk till annan webbplats, öppnas i nytt fönster


Simon Arvidsson
LinkedInlänk till annan webbplats, öppnas i nytt fönster

Mail
GitHublänk till annan webbplats, öppnas i nytt fönster

The number of layers in a forest can tell us a lot about its characteristics. Experts must visit multiple parts of a forest to identify its layers. Can this process be automated?

Please view the video below for a visual commentary.

Layers depend on the height of trees within the forest. Plantations commonly have one layer while natural forests might have three or more layers. Skogsstyrelsen aims to find and protect natural forests, knowing the number of layers could therefore help.

We combine digital records of forest layers with 3D scanned forest areas to train a geometric neural network. This AI model, capable of learning 3D structures, could help experts count the number of layers in a forest without the grunt work.

By evaluating against a baseline, we demonstrate the performance and potential of the network.

The results indicate that geometric deep learning have potential in the field of forestry, but that the problem could be more difficult than expected. The technology could possibly be helpful to find more features than just layering.

Sidan uppdaterad