Data mining by decision tree for object oriented classification of the sugar cane cut kinds

Brazil is the world´s largest sugarcane producer with almost 9 million ha of cultivated area in 2008. Great part of the harvested area is manually cut using the practice of burning the dry leaves prior to the stalk harvest. This practice cause atmospheric pollution and damage to public health, in particular, to local inhabitants. In Sa¿o Paulo State an environmental protocol was signed to establish the burning practice should stop by 2017. Remote sensing satellite images are useful to discriminate different sugar cane harvest types. This study analyzed the generation of decision trees using mean and multi-attributes extracted from objects in TM/Landsat sensor images aiming the classification of types of sugar cane harvesting under different soil types. The classifications performances were between 0.69 up 0.84 kappa indexes. The classifications were sensitives to the different soils and the use of multi-attributes did not contribute to the improvement of the classifications.