Automatic identification of forest species using machine learning methods based on satellite image processing

Monitoring of the condition and species diversity of tree species plays a significant role in the forest resource management. The emergence of high-quality multispectral satellite images opens up opportunities for using information about vegetation in a certain area over a long period of time extracted from images in addition to data obtained as a result of field research. The paper considers the use of satellite images from Sentinel 2 for automated classification using the example of the territory of the natural monument ”Kumysnaya Polyana Nature Park”located in the vicinity of Saratov. Machine learning methods with a teacher were used as automation tools. The paper describes three approaches, differing in the algorithms used, as well as ways of structuring the information obtained from satellite images. As a result of the work, the test part of the study area was classified into seven types of trees, including regions dominated by the most common species. In the future, a similar approach to automating the marking of forest maps will potentially increase the data update interval in field research and increase their effectiveness.

forestmonitoring, Kumysnaya Polyana Nature Park, machine learning, logistic regression, SVM, KNN, decisiontree, algorithm random forest, XGBoost, CatBoost, LightGBM, CNN