Supervised fuzzy classification of SAR data using multiple sources

Synthetic aperture radar images contain a lot of information, but they are difficult to interpret. It is important to extract information dependent on the application to increase the acceptance of radar data. A supervised learning system is necessary to implement the required adaptability. The authors´ fuzzy approach allows one to cope with uncertainties due to linguistic class definition, vague models, mixed pixels and noisy input data. In some cases single-channel classification already leads to satisfying results, however it is often desirable if not necessary to take more than one source of information into account. They show in this paper how the information contained in multiple channels, e.g. the four channels of a full polarimetric SAR and/or information from different sources, can be combined automatically to obtain a final classification with high accuracy. To cope with the problem of different reliability of the available channels as well as in their classification technique the authors make extensively use of fuzzy logic. The proposed methods are computational efficient and increase the accuracy of the classification compared to single channel methods. The performance of these techniques is demonstrated on ESAR data