A Hybrid Approach Classification of Remote Sensing Images

The use of a hybrid approach classification, which combines pixels and objects, has been shown to be suitable for the identification of Landscape Units that contain a variety of land cover objects using VHSR images. However, the pixel-based classification of remote sensing images performed with different classifiers usually produces different results. With the combination of the outputs of a set of classifiers it is possible to obtain a classification that is often more accurate than the individual classifications. In this paper the author analyzes if the use of an output combination of a set of soft classifiers in a hybrid approach classification, can improve the accuracy of the results. To this end, a hybrid classification method was developed that includes the following steps: 1) pixel-based soft classification; 2) computation of the classification uncertainty; 3) development of rules to combine the soft classifications, which incorporate the information provided by the previous pixel-based classification and the results given by the uncertainty measure 4) image segmentation; and 5) object classification based on decision rules which include the results of the combined soft pixel-based classification and its uncertainty. The proposed methodology was applied to an IKONOS image. The overall accuracy of the hybrid approach classification obtained with the proposed methodology was higher than the ones obtained with the individual pixel-based classifications, which shows that the methodology is promising and may be used to increase hybrid classification accuracy.