Neural network learned to identify tree species by satellite


A detailed land use map showing the forest in the state of Chiapas, in southern Mexico. The map was produced using Copernicus Sentinel-2 optical data from April 14, 2016. The image is not part of the study discussed.

Much of what we know about forest management today comes from aerial photos. Whether drones, helicopters or satellites, bird’s-eye views of forests are crucial to understanding how our forests are doing, especially in remote areas that are difficult to monitor in the field.

Satellite imagery, in particular, offers a cheap and effective surveillance tool. But the problem with satellite data is that often the resolution is quite low, and it can be hard to tell what you’re looking at.

Corn a new study using neural networks to distinguish between satellite images can help with this.

Structure of the hierarchical model / Svetlana Illarionova et al., IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing.

“Commercial forest tax providers and their end users, including timber buyers and processors, as well as forest industry entities can use the new technology for the quantitative and qualitative assessment of timber resources in areas rented. In addition, our solution makes it possible to quickly assess underdeveloped forest areas in terms of attractiveness for investment, ”explains Svetlana Illarionova, the first author of the article and Skoltech doctoral student.

Illarionova and her colleagues at the Skoltech Center for Computational and Data-Intensive Science and Engineering (CDISE) and the Skoltech Space Center used a neural network to automate the identification of dominant tree species in high and medium resolution images.

Marking of classes in the study area. Image credits: Illarionova et al.

After the training, neural networks were able to identify the dominant tree species at the test site in Leningrad Oblast, Russia. The data was confirmed by observations on the ground during the year 2018. A hierarchical classification model and additional data, such as the height of the vegetation, allowed to further improve the quality of the predictions while improving the prediction. stability of the algorithm to facilitate its practical application.

The study focused on the identification of dominant species. Of course, among forests of different composition there will be forests where the distribution is roughly equal between two or even more species, but the compositions of these mixed forests were beyond the scope of the study.

“It should be noted that ‘dominant species’ in forestry do not exactly match the biological term ‘species’ and are primarily related to the class and quality of the wood,” the researchers write in the article.

Overall, the algorithm appeared capable of identifying dominant species, although the researchers note that the result may be improved by better training markup, which they plan to do in future research.

“However, in future research we will cover mixed forest cases, which will fall entirely within the hierarchical segmentation scheme. The other goal is to add more forest inventory features, which can also be estimated from satellite imagery, ”the study concludes.


About Author

Leave A Reply