Analysis of Texture Descriptors in Satellital Images to Infer Land Cover Types

An important stage in recognizing Land Cover (LC) on satellite images is the determination of a set of features that can best describe it and help infer between types of covers. This article analyzes and selects texture features of the Gray Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP). The selection of texture features makes it possible to reduce the dimensionality to a relevant subset that provides information in improves the identification between types of LC. This in order to improve performance in semantic segmentation methods in satellite images. The images used for the analysis experiments corresponded to multispectral satellite images from the Sentinel 2 satellite of COPERNICUS of the European Space Agency (ESA). GLCM and LBP features are extracted from patches of different LCs of interest (agricultural vegetation, vegetation, soil, water, and urban). In addition, coefficients of variation of values of the features are obtained, and an analysis of the variation behavior between the different interest coverage and within the same type of LCs is carried out. Finally, relevant features were selected considering the behavior of the coefficient of variation and those obtained with the Relieff method.