Context and objectives
Because of the complexity of urban systems, their spatial and spectral characteristics become more complicated. Affected by the human and natural activities, the urban surface is very heterogeneous and makes the classification of man-made objects in those scenes very difficult using a single sensor with limited spectral range. The goal of this research is to fuse polarimetric SAR and hyperspectral data to classify man-made objects in urban and industrial scenes. While the polarimetric SAR measurements are sensitive to the surface geometry and the dielectric constant of the illuminated surface, hyperspectral data provide information related to the biochemical origin and environment of the observed scene. In May 2004 E-SAR and HyMap data were acquired. The detected images included various man-made materials combining residences, industrial buildings, nuclear power plants, airport and roads in rural-urban and industrial scenes. The E-SAR images are high spatial resolution scene (1.5 m) multi channel (X, L bands), full polarimetric (L-HH, VV, VH, HV) and interferometry (single path) data. The HyMap images are 4 meters spatial resolution data collected using 126 spectral bands.
Expected scientific results
Generally, we have found that data fusion of SAR- and hyperspectral data give complementary information regarding the urban scene and are useful for road detection. However, data fusion of E-SAR and hyperspectral was not sufficient for building detection due to different imaging geometry. Refined methods have to be developed and applied for future improvements. Moreover, we have found that “point scatterers” detected in SAR images are strongly correlated with man-made structures signals. For hyperspectral data it is possible to rely on a pixel-wise classifier for extraction of roads, buildings and building materials. For SAR images a pixel-wise classifier can be used for large buildings or for identifying built-up areas. However it is not possible to detect individual houses in an urban environment using SAR, mainly due to the imaging geometry. A classifier based on polarimetric decomposition methods is useful for detection of large roads in SAR. However, a specific detector for linear feature is more effective for smaller roads.