Processing of Satellite Images of Images Using Software

Abstract

Remote sensing of ground coverings requires the application of robust classification methods that allow accurate determination of the ground cover of study areas, and in many cases very complex. In recent years, they have proposed a large number of classification methodologies, a large part of which depends on artificial intelligence. The most well-known include Decision Trees (AD), artificial neural networks (RNA), machine support vectors (MVS) and a group of compilers such as Random Forest (RF). This study aims to study how the selection of a work affects the accuracy of the maps and analyzes the statistical significance of the differences in the results obtained by the different classification methodologies. The different classification methods were assessed quantitatively, taking into account 600,000 different sets of parameters for the classification of land cover in the province of Granada. To classify the study area, 14 classes of surfaces were identified. Different soil and Landsat TM5 images were used for the spring and summer seasons as well as various auxiliary variables obtained from the digital terrain model. Classification techniques RNA, MVS and RF yielded maps greater than 91% accuracy. M behaved worse than the rest of the methodologies. However, the results of this study show differences. Important in the efficiency..