07 August 2023
New sensor technologies and remote-sensing platforms brought high-resolution spatial and temporal data to forest monitoring. At the same time, these are huge volumes of data that need to be processed before allowing for tasks such as species classification and damage assessment, or even pattern identification for predicting areas at risk for different threats.
In a work presented at the 20th International Conference on Smart Technologies (IEEE EUROCON 2023), held in Torino, Italy, in July 2023, SWIFTT partners from the Space Research Institute of Ukraine and collaborators deal with the problem of semantic segmentation - the identification and classification of collections of pixels according to various characteristics - of satellite imagery to deliver high-resolution forest type maps.
They propose 4 machine learning models, two based on Random Forest (RF) and two based on Convolutional Neural Network (CNN) - U-Net. Using as input 2 Sentinel-2 images of (one for summer and one for winter, 4 spectral bands from each), and as output (labels) the Copernicus Forest Type dataset for 2018, the authors’ models showed promising results on validation data. Among them, one based on U-Net ended up being the most efficient in forest classification with overall accuracy 91.7%. At the same time the best RF model scored 86.5%.
In order to check for model transferability, the authors created and compared forest maps of the northern part of the Kyiv region from 2018 and 2022. The experiment confirmed the robustness of the model and its scalability.
According to the authors, the developed map can provide valuable data for foresters, biologists, or other researchers to make decisions about forest management and conservation, and to ensure that European forests are managed in an ecologically sustainable way.
Source: Y. Salii, V. Kuzin, A. Hohol, N. Kussul and H.
Yailymova, "Machine Learning Models and Technology for Classification of
Forest on Satellite Data," IEEE EUROCON 2023 - 20th International
Conference on Smart Technologies, Torino, Italy, 2023, pp. 93-98, DOI: 10.1109/EUROCON56442.2023.10199006