EVOLUÇÃO DAS TÉCNICAS DE CLASSIFICAÇÃO DE IMAGENS DO SENSORIAMENTO REMOTO UTILIZADAS NA PRODUÇÃO CIENTÍFICA BRASILEIRA

Evolution of remote sensing image classification techniques used in Brazilian scientific production

Abstract

Currently, the images interpretation is based on automatic classification. This is subdivided into several methods, each one based on a type of algorithm. Thus, it is important to identify the classification techniques most used in Brazilian scientific production, as well as their respective results and their evolution over the years. In order to guide future works that intend to use these methods as well as identify trends in their applications. This work presents a bibliometric analysis of the production of Brazilian theses and dissertations about automatic image classifiers as well quantifying as analyzing the most mentioned classification methods. The research was carried out in the BDTD (Brazilian Digital Library of Theses and Dissertations) database, to obtain the main characteristics of the methods employed. The results showed that the most mentioned automatic classifiers, in absolute numbers in descending order are: Maximum Likelihood, Neural Networks, Support Vector Machine (SVM), Decision Trees, Random Forest, Minimum Euclidean Distance, Spectral Angle Mapper (SAM), K-Nearest Neighbors (KNN) and ISODATA. It was concluded that the algorithms are linked to the Machine Learning concept are on the rise in numbers of more recent publications, and that the Decision Trees and Neural Networks classifiers present the highest accuracy values ​​in the analyzed works.