Robust classification of targets in POL-SAR using wavelet packets

Some polarimetric SAR (POL-SAR) platforms have indicated the promise of producing imagery of a scene acquired at several frequencies, incidence angles, polarizations, and at multiple time intervals. Such large amounts of data necessitate the construction of robust algorithms that can process a wide range of data with minimal retraining and supervision while minimizing the complexity of the algorithm. For statistical target detection and classification, the challenge is to build a robust classifier that operates satisfactorily even when the statistics of the desired targets migrate from a single prescribed model. We present a technique for performing robust target classification when the targets are different terrain types. The algorithm accounts for variability in terrain signatures by deriving a single representative process for each terrain from a family of stochastic scattering models. A best-basis search through a wavelet packet tree, using the Bhattacharyya coefficient as a cost measure, determines the optimal unitary basis of eigenvectors for the representative process and offers a scale-based interpretation of the scattering phenomena. The associated eigenvalues and means are determined through iterative algorithms. The technique is tested using images acquired from the BOREAS field campaigns in Canada.