Post-segmentation feature-based classification of synthetic aperture radar data

Classification is one of the most important image analysis tasks as it provides image data with labels that transform it into information about the real world. In this research, classification of clutter in single and multichannel very high-resolution airborne SAR imagery is achieved by extending several classification methodologies to overcome limitations which exist when applied to SAR data. This is achieved by first pre-processing the data using segmentation techniques, then classifying the resulting regions using a number of features calculated over the regions. By attempting to classify regions as opposed to single pixels we can increase the dimensionality of the feature space even for single channel data, and overcome the problems of speckle and texture inherent in SAR imagery. Both supervised and unsupervised classification methodologies have been adapted to feature-based region classification. The methodology has been extended to multichannel data, specifically polarimetric data. Examples of the application of the proposed method are given for several very high-resolution data sets from the QinetiQ airborne SAR. It is concluded that the proposed approach provides a successful and flexible, sensor independent solution to the problem of classification of very high-resolution SAR data for many different applications.