The emerging field of imaging spectroscopy is a powerful new technique for use in the surface reconstruction problem. In it, highly-dimensional spectral data are analyzed using vector-based strategies which support object space classification at the single pixel level. In practice, image space radiance data are usually inverted to recover estimates of object space reflectance, and these reflectance spectra are matched against a library of known object space spectra. Consequently, the accuracy of the inversion from radiance to reflectance is important. In this research, we explore a novel approach to this inversion.; We treat imaging spectroscopy as an approximately Linear Shift-Invariant (LSI) System and approach the inversion problem as a Multi-Channel Restoration (MCR) computation. We call this approach Radiance Refinement by Multi-Channel Restoration (RRMCR). RRMCR has the potential to simultaneously correct perturbations to radiance which are induced by the atmosphere and the sensor itself, and specifically address the radiance bias introduced by the well-known adjacency effect.; In this research, we report on the implementation of RRMCR using both a priori and a posteriori system identification strategies. We first deconvolve the calibrated Point Spread Function (PSF) from each channel of a hypercube using inverse filtering in the frequency domain. We then conduct a channel-by-channel estimation of the system PSF using Edge Gradient Analysis (EGA), and re-compute the inverse filter for the hypercube.; In these experiments, we use a novel dual-altitude approach wherein we employ a low-altitude image as a proxy for the object space radiance data. Specifically, we restore the higher altitude radiance data to the radiance measured at a lower altitude. The two airborne images are acquired with short temporal separation such that irradiance and atmospheric conditions are assumed to be constant. This strategy allows us to avoid issues associated with the inter-calibration of ground level and airborne spectrometers.; Our work demonstrates the potential of this approach for removing a radiance bias caused by the known across-track over sample in a specific spectrometer, the Airborne Visible and Infrared Imaging Spectrometer (AVIRIS). We also show that the RRMCR approach can be used with an EGA-derived PSF to compensate for total system perturbations, though this computation tends to increase the noise in the radiance data and requires that we exercise great care in the system identification step.
展开▼
