Land Use Classification Using Optimum-Path Forest

It was introduced in this paper the Optimum-Path Forest for land use classification aiming a better environmental management, using images obtained from CBERS 2B CCD satellite covering the area of the Rio das Pedras watershed, Itatinga City, São Paulo State, Brazil. We also compared the Optimum-Path Forest algorithm with the well known supervised classifiers: Artificial Neural Networks using Multilayer Perceptrons, Bayesian Classifier and Support Vector Machines. The Optimum-Path Forest and Support Vector Machines classifiers had similar results and outperformed the other ones, but the first was much faster than the last one. As far we know, we are the first that used the Support Vector Machines classifier in this context of research field. Also were presented some qualitative results, in which the Optimum-Path Forest and Maximum Likelihood classifiers were compared against each other for visual purposes.