Multisource classification based on uncertainty maps

A new methodology to perform classification of multisource data is proposed in this work. The technique takes into account the classification results (a classified image and an uncertainty map) derived from each data source in order to generate a new classified image. It is based on a region-based classifier which employs stochastic distances, statistical tests and reliability of the classification to produce a final classification. Images acquired from two distinct and independent data sources (optical and microwave sensors) were used to validate the proposed methodology. One LANDSAT5/TM image and one ALOS/PALSAR image from a region on Brazilian Amazon were classified, firstly individually and then using the proposed classification technique. The results showed that the individual classification results can be improved by the use of our multisource classification technique. It can be concluded that this new method to combine information derived from different data sources is strongly promising.