The role of frequency and polarization in terrain classificationusing SAR data

摘要

The expected accuracies of land-cover classification are evaluatedfor existing and potential orbital SAR systems. Land-coverclassifications are compared for ERS-1, JERS-1, SIR-C and X-SAR. Inaddition, SIR-C/X-SAR data from a largely forested test site in northernMichigan are used to simulate the expected performance of potentialorbital SAR systems such as Envisat, PALSAR and LightSAR. Theclassification approach uses orthorectified and filtered SIR-C/X-SARdata overlain with known polygons subdivided into spatially independenttraining and testing populations. For each potential sensorconfiguration, the relevant feature vectors are subsampled for a portionof the image and used to generate unsupervised clusters. These clustersare then assigned to the known classes of the training population usingmaximum likelihood criteria with equal probabilities. Contingency tablesare produced for the testing population using minimum distance criteria.The classification results show that longer wavelengths (such as L-band)are of greatest value for discriminating general land-cover classes onthe basis of biomass and roughness since there is a greater dynamicrange relative to these attributes. Shorter wavelengths (C-band orX-band) are more sensitive to smaller scattering elements such asfoliage and small stems and are therefore of importance indiscriminations related to crown-layer architecture (i.e., leaf size andshape). The best results are achieved when classification is based uponmultiple frequency data