MAPEAMENTO DE ÁREAS PERMEÁVEIS INTRALOTES URBANAS A PARTIR DE IMAGENS OBTIDAS POR VANTS PROCESSADAS POR GEOBIA E MINERAÇÃO DE DADOS

Abstract

Disordered urbanization and the resulting changes in land use have resulted in the recurrent occurrence of urban flood events, demonstrating the need for improvement in urban planning and expansion policies. In this context, the monitoring of land use and occupation through geotechnologies is an important support tool for decision making by the public administration. With the advances in technology, the availability of high spatial resolution images has become popular, however, this increase in spatial resolution requires an automatic classification approach in which pixels are not individually classified. The classification of object-based images, in this work treated by the term GEOBIA (Geographic Object-Based Image Analysis), presents itself as one of the most efficient techniques in the classification of high spatial resolution images. However, the costs associated with purchasing commercial software licenses with this approach are usually high. Therefore, this research aimed to evaluate the effectiveness of using GEOBIA and data mining techniques developed in free software applied to aerial images of very high spatial resolution UAVs imagery, as a tool for monitoring and inspecting constructed areas at parcel levels. For this purpose, two orthomosaic were used: one with a spatial resolution of 3 cm and the other of 6,2 cm, both of which are study areas located in the urban area of Irati-PR. Semiautomatic classifications were developed using the public domain software TerraView and its GeoDMA plugin, using the MRS segmenter and data mining with decision tree algorithms. When evaluating the results obtained considering 11 predefined land cover classes, and applying the manual classification by vectorization on canvas as ground truth, the Kappa values (0,600 and 0,584), indicated a “Moderate” classification in both study areas. While when grouping some classes of land cover spectral similarly, the Kappa values (0,685 and 0,685) went up and started to indicate “Substantial” classifications. When analyzing the land permeability classifications applied to determine the target parcel constructed area in this study, the Kappa values found (0,762 and 0,782) also indicated a “Substantial” classification for both study areas. In addition, when applying the t-Student hypothesis test considering a 95% confidence level, it can be concluded that in neither of the two study areas there was a significant difference between the constructed areas obtained with the manual classification and those obtained based on GEOBIA techniques. Thus, the results demonstrate the potential of the proposed methodology for obtaining data from very high spatial resolution, UAV imagery.