Classification of polarimetric SAR images of suburban areas using joint annealed segmentation and H/A/α polarimetric decomposition

In this paper, a joint segmentation-based classification technique is proposed for fully polarimetric SAR images of suburban areas. This is based on the joint use of the H/A/alpha polarimetric decomposition and multivariate annealed segmentation. The introduction of a segmentation stage before classification allows the exploitation of the information available in the noisy estimate of the anisotropy parameter, retaining at the same time the highest possible spatial resolution. Specifically, in the joint segmentation of entropy, angle and anisotropy images, the latter is exploited when it provides a significant contribution, but the process is mainly driven by the more stable parameters (entropy and angle). When jointly classifying all the pixels belonging to the identified homogeneous segment, the anisotropy parameter can be effectively exploited without introducing any additional noise in the classification output. The impact of the exploitation of the anisotropy parameter is investigated quantitatively through the application of the proposed technique to a AIRSAR C- and L-band image of suburban areas. To characterize the limits that can be achieved, both lower and upper bounds to classification performance are introduced, corresponding respectively to pixel-based classification and joint classification of all the pixels in the regions defined by the ground truth. Results show that a significant performance improvement can be achieved by classifying the homogeneous regions identified by segmentation, instead of single pixels or even small windows of 3 x 3 pixels. Moreover, it is shown that the exploitation of the anisotropy parameter allows a better classification accuracy, and in particular a better discrimination of built-up areas from other classes. (C) 2003 Elsevier Science B.V. All rights reserved.