Spectral attributes selection based on data mining for remote sensing image classification

Remote sensing images are a rich source of information for studying large-scale geographic areas. The new satellite generations have producing huge amounts of data. Data mining techniques have been emerged last years as powerful tools to help in the analysis of these data. In the area of remote sensing image analysis, software like GeoDMA, eCognition, InterIMAGE, and others are available for end users. These software provides tools to extract several attributes of the images. These attributes are then used in image classification and analysis. When dealing with high resolution multispectral satellites, we have a large quantity of attributes. In many cases, the attributes are highly correlated, and consequently may not help to separate the classes of interest. Thus, this work shows the results of an approach to analyze the correlation of the attributes between several classes of interest, selecting those that will better distinguish them. In this way, it is possible to reduce the amount of data to be used during classification and analysis, consequently reducing the computational time for classification.