This work investigated the adherence of data mining techniques oriented to data classification problems in the identification of sugarcane crop areas in Landsat 5/TM images. To do so, pixels of images having sugarcane crop areas were studied in three different phenological phases. Such pixels were converted into surface reflectance values in neighborhood of the towns Araras, Araraquara and São Carlos in São Paulo State. It were generated five decision tree models using the algorithm C4.5 and all of them produced accuracy rates above 90%. The introduction of texture attributes brought significant gains in accuracy of the classification model and helped improve the model of distinction of areas cultivated with sugarcane in the midst of various types of land cover, such as bare soil, urban areas, lakes and rivers. The vegetation indices were relevant in distinguishing phenological phases. The results support the potential of decision trees in process of classification and identification of areas cultivated with sugarcane in different cities inside São Paulo state.
agricultural mapping; image classification; decision tree; remote sensing
