Comparação de métodos de classificação de imagens digitais na detecção de alvos indicadores de impacto ambiental em linhas de transmissão de energia elétrica

Abstract

The projects in the energy sector, related to transmission lines, are composed of cables and metal structures, although providing benefits, the installation and operation still have significant environmental impacts along its stretch of passage. In this context, the aim of this study is to identify areas of "soil exposed", which may be related to erosion processes, and "fragments of native vegetation" in the range of direct influence of power transmission lines through computational techniques and remote sensing data. Tests were lead on Orthophotos, using the Maximum Likelihood Classifiers and Neural Networks, as well as the Object-Oriented Segmentation, integrating the Orthophotos with Laser Scanner data. The results analyzed based on Commission errors (accuracy of the user), Omission (accuracy producer) and kappa index. According to Kappa index calculated, the results showed variations between moderate and excellent content, indicating significant results with regard to the methodology used. Referring to the Commission and Omission errors, disarray in the classification process took place between classes presenting pixels with similar digital values within the same scene, especially within the classes that have different types of vegetation. With respect to the environmental diagnosis found in the range of direct influence of the transmission line, the portion of the sample median, indicates a high rate of anthropogenic interference. The class "Campo Limpo" had greater prominence, especially in areas with agricultural activities. The areas directly affected need attention as their recovery. It is emphasized that the results, which are found, is the first effort of diagnosis and research, requiring occasional visits in loco to prepare the planning recovery goals.