ARTIFICIAL INTELLIGENCE AND ORBITAL IMAGES APPLICATION FOR ANALYSIS OF SPATIAL LAND USE AND COVERAGE PATTERNS

Abstract

The study aimed to analyze the performance of different machine learning (ML) algorithms in predicting land use and land cover patterns from time series spectral data from Thematic Mapper (TM) and Operational Land Imager (OLI) sensors. The QGIS software was used, where the import of TM / Landsat 5 images began in 2004 and 2009 and OLI/Landsat 8 for 2015 and 2019, to obtain information to characterize and differentiate usage patterns and land cover. Subsequently, training and testing of the algorithms, Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (K-NN), and Naive Bayes (NB), were carried out in the proportions of 80%-20%, 70%-30%, 60%-40% in the KNIME software. The performance was analyzed based on global accuracy and the Kappa index. The RF and SVM for the years 2004 and 2009 showed the best performance (global accuracy), while for the years 2015 and 2019, they were the K-NN and the RF. The Kappa index values indicated that the classifications of the algorithms varied from 0.80 – 1.00. The proportion of 60% (training) and 40% (test) was the one that provided the best results for all the dates analyzed. The data from the pixels sampled from the land use and land cover patterns of the TM and OLI sensor images proved to be efficient for the ML process in the KNIME software.