Study On Sea Ice Classification Using Multi-frequency Polarimetric SAR Data

Sea ice classification is a very important application in the field of sea iceSAR remote sensing monitoring. Multi-frequency polarimetric SAR data, comparedto the traditional single-frequency SAR data, could provide more information suchas polarization, phase and energy in different frequency, which allows for a morecomprehensive description of the very complicated electromagnetic scatteringcharacteristics of sea ice and enhances capabilities of SAR data for sea iceclassification and identification. In this dissertation, C-and L-frequencypolarimetric SAR data obtained from spaceborne SIR-C SAR was used on anin-depth study on the application of classification and identification, major worksand reads as follows:1) This dissertation illustrated the basic theory of polarimetric SAR andsummarized two major categories polarimetric SAR quantities commonly used inthe classification and identification classification of sea ice. The extraction andrepresentation method, origin and mutual relations of those quantities were alsointroduced.2) Based on C and L-frequency polarimetric sea ice SAR data acquired by theSIR-C SAR system in the Eastern Weddell Sea, Antarctica，a sample statisticalmethod and a supervision classifier were both used to make a qualitative andquantitative evaluation of sea ice recognition ability of single-frequency SARpolarimetric quantities as well as direct combined dual-frequency SAR polarimetricquantities.Based thereon, the benefits and limitation of single-frequency as well asdirect combined dual-frequency SAR polarimetric quantities for sea iceclassification and identification in the Eastern Weddell Sea, Antarctica, areinvestigated.3) For the limitation of direct combined SIR-C dual-frequency SARpolarimetric quantities for sea ice classification, two data fusion methods wereintroduced to fuse dual-frequency SAR polarimetric quantities to further enhancetheir ability in sea ice classification and identification. The maximum likelihoodclassifier is used to verify the ability of the fusion data of sea ice classification. Theexperimental results showed that the principal component analysis (PCA) caneffectively improve the ability of SIR-C dual-frequency SAR polarimetric quantities in sea ice classification, while the fusion data acquired by locally linearembedding (LLE) algorithm did not seem to help to improving the effectiveness ofdual-frequency SAR polarimetric quantities in sea ice classification.