Machine learning for forest segmentation in satellite images

Download article in PDF format

Authors: Mihaylov E. V., Say S. V.

Annotation: A method to segment forest regions in satellite images is pro- posed. Satellite image analysis usually includes computation of NDVI to segment vegetation areas based on spectral information from red and infrared channels. To segment forest and grass regions we use texture information, as forest areas usually include a lot of small shadows. In the article only simple texture features are considered, such as standard deviation and skewness. Instead of manual selection of thresholds for each feature, we use machine learning to classify regions

Keywords: satellite image processing, textural features, classification, machine learning