Reconstruction Algorithm Characterization and Performance Monitoring in Limited-Angle Chromotomography

摘要

Hyperspectral data collection and analysis is an increasing priority with the growing need to obtain greater classification precision than offered by traditional spatial imagery. In this thesis, trends in hyperspectral chromotomographic reconstruction are explored where reconstruction is performed using a series of spatial-chromatic images. Chromotomography involves capturing a series of two-dimensional images where each image is created by placing a prism in front of the focal plane array; causing spectral dispersion corresponding to a series of prism angles over a single rotation. Before testing reconstruction, synthetic data is produced, approximating what would be produced from prism dispersion on the focal plane array. The pseudo-inverse singular matrix problem is addressed where two methods are compared to find which produces minimal error. The standard iterative error reduction algorithm, SVD- POCS, is shown to be incapable of reconstructing the mean of the source scene, making absolute radiometry analysis impractical. However, SVD-POCS is shown to provide the least error if the goal is to perform relative radiometry analysis. Additional constrains are needed to make absolute radiometry analysis possible. The added constraints of non-negativity, spatial extent of the cold field stop, forcing the sum, and keeping the mean for each iteration improves absolute radiometric performance. These additional constraints also allow use of a warm field stop to monitor reconstruction error for both the pseudo-inverse and iterative improvement algorithm. Error can be calculated each iteration to ascertain when a minimum has been reached in a mean square error sense. Thus, minimum mean square error of the reconstruction can be obtained with confidence.