Spatial Properties of Forest Clutter Measured by Wavelet Frames in Polarimetric PALSAR Imagery

The paper reports on experimental analysis of the spatial statistics (texture), as measured by wavelet frames in Lband SAR observations of tropical forest. Signatures are used to characterize the dependences of the texture measures on scale and polarization state. For detected data, experiments performed using a full resolution (12.5 m) JERS-1 image confirm that textural features provide good separability for two forest classes of interest (swamp and lowland rain forest). However, spatial resolution turns out to be a key factor influencing separability in the texture feature space. Indeed simulation of a PALSAR strip image at 50 m pixel spacing indicates that this product does provide textural information only at coarse resolution (800 m). Concerning polarimetry, analysis of a fully polarimetric PALSAR scene reveals that thematic classes such as primary forest, swamp (flooded) forest, degraded forest are uniquely characterized by texture signatures in the scale-polarization plane.