Monday, June 21, 2010
Ubuntu on Hawkboard and Statistics on SAR
The mind is a massively parallel device - at least at the neuron level. I am always doing multiple things simulataneously. Multiple in the sense of intellectual heritage, yet unified in a sense that I am doing something that interests me and I might as well call it an unique discipline - how to see at all parts of the spectrum or how to get to mars. In essence it is domain unified.
Last few days I have been looking at arm twisting the low-power DSP on the hawkboard into doing some image processing. Porting OpenCV to the DSP on beagleboard seems to be proceesing pretty well. I will have to build that and give it a try. For now I am satisfied setting up a dsplink enabled kernel and Ubuntu image from an SD card. The root image does not contain UVC drivers so I can't use my USB cameras, but at least the Hawkboard is operational. Time to switch attention to the BeagleBoard and give the dsp enabled opencv a whirl.
On the otherside I did some eigen decompositions of dual-polarimetric TerraSAR-X and PALSAR and noted how wildly different grasslands appear at L-band vs X-band. While L-band sees predominantly surface scattering, X-band is carried away by volume scattering. The X-band VV channel seems statistically most responsive to the pasture biomass, but the confidence of my statistics with only 33 sample points is rather rudimentary. The pastures are fairly dry (10% moisture) as opposed to wheat I have worked with in the past (50-60% moisture), this significantly reduces the radar cross-section and the net backscattering needs to be explained in terms of additional parameters. In this case there is significant influence from surface roughness and moisture content, backscattering increasing with increasing roughness and decreasing with increasing moisture content. Where the entropy is low I will investigate the negative-alpha filter for retrieving soil moisture.
In dual-polarimetry the alpha angle contains ambiguity and can often be misrepresented as multiple scattering. This ambiguity cannot be resolved without considering the actual targets of interest.