Classificação de Imagem Orbital Rapideye utilizando banco de dados NOSQL e método GEOBIA

Abstract

The information from images captured by Remote Sensing and the techniques available in the Geographic Information Systems, it is possible to generate thematic mappings for use and land cover. For this, the classification of images is realized to define interest classes. This classification can be done pixel by pixel or by regions. In high resolution images, such as Rapideye, classification by region is indicated. This method considers the information of the pixel and its neighborhood, grouping pixels with similar characteristics create the regions. Therefore, it is recommended to apply the GEOBIA segmentation method, which segments the image in regions to extract spatial, spectral and texture characteristics. As a result of this method, have the region vector and the relational database with the attributes (spatial, spectral and texture). The objective of this work was to obtain the classification of the use and coverage of the soil of the Rapideye image using the NoSQL database oriented to graphs to analyze the attributes extracted through GEOBIA. The developed methodology used the Multivariate Analysis to analyze the attributes resulting from the segmentation. The dendrogram it was possible to separate the groups of attributes (spatial, spectral and texture), which were used for the search queries by groupings of regions with similar characteristics in the graph formed by the NoSQL database. The regions were classified according to the interest classes defined in the photointerpretation process, generating the classified image. To validate the result, the image area of the study area was classified by the Minimum Distance, Maximum Likelihood and KNN algorithms and the confusion matrix. The KNN algorithm presented better classification, with a kappa index of 0.77 and was then used for comparison with the image classified by the NoSQL database, through cross tabulation. The cross-validation of the data showed that the image classified by the NoSQL database obtained positive results. It was concluded that the research reached the proposed objectives presenting satisfactory results for the method developed for classification of land use and land cover.