A COMPARISON OF 2ND-ORDER CLASSIFIERS FOR SAR SEA-ICE DISCRIMINATION

In this paper we present results of an analysis of the relative utility of statistical, structural, and frequency based second-order texture methods for discrimination of sea ice types in synthetic aperture radar (SAR) data. Algorithms were trained using a calibration data set and robustness of the methods were assessed by directly computing ice classes within a validation data set Classification using a first-order approach (average grey level) produced Kappa classification accuracies of 51.0 and 33. 0 percent for the calibration and validation data. The first-order approach is provided primarily as a reference from which to compare the second-order approaches because the test conditions were selected to be specifically difficult (i.e., different incidence angle ranges between calibration and validation images) for any approach using image tone or the relative scattering cross section. Results from the second-order approaches indicate that the two spatial domain statistical approaches, Grey Level Co-Occurrence Matrix (GLCM) and the Neighboring Grey Level Dependence Matrix (NGLDM) provided high classification accuracies under the difficult test conditions examined here. The GLCM results achieved a Kappa Coefficient of 84. 0 and 81.0 percent for the calibration and validation sets. The NGLDM achieved a Kappa Coefficient of 83. 0 and 76. 0 percent for the calibration and validation data sets. These results are statistically equivalent between the calibration and validation data sets and between the GLCM and NGLDM schemes. The Spatial/Spatial Frequency (S/Sf) approach appears to be sensitive to the training conditions generated from the calibration data set and therefore do not provide statistically reproducible results between the calibration (87 percent) and validation (18 percent) test conditions. Results from the Primitive Texture Value (PTV) method suggest poor operational capabilities both due to low calibration (65 percent) and validation (58 percent) accuracies.