Earth remote sensing data processing for obtaining vegetation types maps

In this paper, we propose an earth remote sensing data processing technology for obtaining vegetation types maps. The technology includes the following steps: obtaining superpixel representation of an image, calculating superpixel features, K-Means clustering of superpixels by a user-defined training sample, and obtaining vegetation types maps. When compared to other solutions, the major difference of the proposed technology is the ability to combine superpixel segmentation and feature calculation into a single process in one pass of an image that reduces the computational complexity. Another difference lies in the way of forming a sample dataset using superpixel representation of an image. The advantages of the proposed technology are the use of a smaller training dataset and a higher classification quality in comparison with the elemental classification.