Hyper-spectral sensor calibration extrapolated from multi-spectral measurements.

摘要

Hyper-spectral (HS) sensors are the instruments of choice for remote sensing applications involving environmental monitoring, littoral survey, and military assessment. Accurate band-to-band sensor radiometric calibration is critical for successful data mining of such HS spectral sets. Current calibration is often performed with methods not necessarily developed for HS applications. This work describes two advances which facilitate laboratory source calibrations. First, an analytical solution to the attenuation of flux within an integrating sphere, the best laboratory source of non-directional radiance for numerous radiometric applications, is given. Relative component attenuations due to integrating sphere coating, exit port escape, and atmospheric absorption are derived employing a geometrical PDF of summed probabilities. Equations providing the attenuation ratios and mean number of reflections for the three outcomes are obtained, yielding the three partial mean pathlengths and variances of all quantities. This work then describes an approach allowing accurate radiometric calibration of HS sensor bands using well-characterized and stable multi-spectral transfer radiometers. The resulting high-quality calibration enables the best representation of the truth spectral signature of the imaged scene. In order to obtain the best calibration with the least instrument complexity and expense, it is critical that the radiometer samples the source with the fewest samples at those optimal wavelengths which predict that source with the highest accuracy. The optimal source-specific bands are determined efficiently by application of the Direct Search methodology described here. Using the minimal selection of multi-spectral radiometer measurements obtained from the optimized transfer radiometer bands, one can obtain a complete and accurate calibration set for the continuum of calibration coefficients required for a robust HS application. Degradation of the prediction is documented for several typical error sources encountered with calibration, thereby defining limitations on the usefulness of the optimization approach.