Unsupervised classification of scattering mechanisms in polarimetric SAR data using fuzzy logic in entropy and alpha plane

The eigenvalue-eigenvector-based approach for understanding the scattering mechanisms of polarimetric synthetic aperture radar (POLSAR) data leads to noisy classification results due to arbitrarily fixed zone boundaries in the H/alpha plane. In this paper, a new classification scheme that can address the inherent vagueness of class boundaries in the H/alpha plane was tested in order to improve the unsupervised classification of the microwave scattering mechanism by introducing concepts related to fuzzy sets. A 2-D fuzzy membership function was developed for the fuzzification of the 2-D H/alpha plane. The proposed fuzzy H/alpha classifier is composed of three steps: fuzzification of the H/a plane, iterative refinement of membership degrees using the c-means algorithm, and defuzzification for the final decision process. The performance of this new approach for the L-band NASA/Jet Propulsion Laboratory's Airborne SAR data obtained during the PACRIM-II experiment was shown to be consistently improved. This new classification technique can be applied to POLSAR data without any a priori information. The fuzzification of the zone boundaries can be further applied to the interpretation of the POLSAR data, e.g., multifrequency classification, retrieval of bio- and geophysical parameters, etc. In order to propose another implementation of the fuzzy boundary representation, we exploited the combination of the H/alpha state space and anisotropy information.