Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2012 · 29 citations · 25 references
ComputingComputer ScienceEarth SciencesMachine LearningMaterials CharacterizationRemote SensingImage ClassificationPolar Remote SensingPolarization OpticsPolarization-based ImagingRemote Sensing ImagerySupport Vector MachinesPolarimetric Feature ExtractionPolarimetric Remote SensingReceiver Operating Characteristic AnalysisRemote Sensing AnalysisRemote Sensing InstrumentationRemote Wave SensingSpectral ClassificationTransformations
Active (Mueller matrix) remote sensing is an under-utilized technique for material discrimination and classication. A full Mueller matrix instrument returns more information than a passive (Stokes) polarimeter; Mueller polarimeters measure depolarization and other linear transformations that materials impart on incident Stokes vectors, which passive polarimeters cannot measure. This increase in information therefore allows for better classication of materials (in general). Ideally, material classication over the entire polarized BRDF is desired, but sets of Mueller matrices for dierent materials are generally not separable by a linear classier over elevation and azimuthal target angles. We apply non-linear support vector machines (SVM) to classify materials over BRDF (all relevant angles) and show variations in receiver operator characteristic curves with scene composition and number of Mueller matrix channels in the observation.
25
Corinna Cortes, Vladimir Vapnik · Machine Learning · 1995 · 33K citations · Full text
A tutorial on support vector regression
Alex Smola, Bernhard Schölkopf · Statistics and Computing · 2004 · 13K citations