Monitoramento das áreas cultivadas de soja e arroz através da classificação de imagens orbitais.

This work aims at introduce a new cognitive classification method based on color attributes. Implemented in a supervised manner, the method assumes the user collect samples from unlimited classes in a limited space of attributes (only three). Then, the samples elements are converted to HSV space and plotted in a reduced HSV colored diagram. The user has also to select polygons on this HSV space in order to generalize the comprehensive space of each class. Finally, the image is all converted to HSV space and each element is considered in order to define if it lies in a region occupied by one class in the HSV reduced space. The main advantage of the proposed classification process is the power to emulate the human knowledge used by photo interpreter during visual interpretation of remote sensing image targets. Experiments performed with an image marked by deforestation in Amazon were conducted by comparing the performance of several classification approaches. The method proved to be useful in noncomplex problems when simple approaches tend to show adequate results. Other advantages of the proposed method is its simplicity and interactivity, besides it ability to generalize the sampling process when taking homogeneous samples is a difficult task.