Automatic Identification of Tree Species in Arboreal Forest

There is significant variation in our forests. This becomes evident when I look at the tree images captured by users measuring tree diameters with the Arboreal Forest app.

Almost two years ago, we launched automatic tree species identification in the Arboreal Forest app for the Scandinavian countries. We had taken numerous pictures of trees near my home and trained an AI model to recognize Pine, Spruce, and Birch. When I tested it myself, it worked well, but I realized that the captured images included areas beyond the tree trunk. This created a problem where the background could influence the AI's species identification. Last autumn, we modified the app to use less of the background for species identification.

Image: Example of different tree species,. Birch and three spruce

In the app, you can activate "Save Images" - this saves images, and we at Arboreal gain access to these images. At the end of June 2022, I downloaded several thousand of these images of pines, spruces, and birches.

I first tested how the old AI model performed in identifying 538 of these images. It worked worse than expected. Several images that should have been correctly identified were misclassified.

Therefore, I trained a new model that performed better than the old one. I used images that you, the users, had collected. It's still not perfect; it has the most difficulty recognizing birches with a lot of blue-green algae and bark texture, as shown in the example below. It sometimes also fails to recognize spruces, but it's very good at identifying pines.

Image: Trees that are hard to identify for an AI. From left, tree with samge, tree with a lot of lichens and a tree with snow

This improvement in automatic tree species identification enhances the Arboreal Forest app's capabilities, making it more accurate and reliable across diverse forest environments. We continue to refine our AI models based on real-world data, ensuring that the app evolves to meet the needs of foresters and researchers in the field.

Your continued use of the app and image contributions play a crucial role in this development. By activating "Save Images," you're helping us improve the technology for everyone. We appreciate your support in making Arboreal Forest an increasingly powerful tool for forest management and research.

This is mainly due to the fact that we have relatively few images of birches, as coniferous trees are much more common in our forests. As we get more birch images, we will retrain the model and achieve better results.

Feel free to activate the automatic image recognition feature and let us know how it works for you.