Using Neural Networks and Machine Learning Methods to Detect Clearcut Regions in SENTINEL–2 Satellite Imagery

At the moment, using satellite images for environmental monitoring is the most advantageous and convenient solution. Without the use of modern technologies, to find the illegal cutting sites in hard-to-reach forest areas seems almost impossible. Timely detection of damaged areas will help reduce the risks of deforestation. This will preserve the natural habitat of many animal species; help prevent a decrease in oxygen production and control natural resources. In other works, experts use convolutional neural networks, as well as models such as RandomForest, XGboost and other less popular ones. In this article, the authors compare the effectiveness of such methods using different metrics. This work conducted a study using Sentinel-2 satellite data sets in geoTiff format, divided into $256\times 256$ pixel fragments. To find the most accurate way to detect anomalies in forest images, machine learning methods such as Random Forest, XGBoost, and the U-Net convolutional neural network were applied.