nai

Project Progress

In the first year we have been developing algorithms for orbital data analysis, feature extraction and novelty. In particular we have refined a process of registering orbital multispectral data from ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) to geographic locations at which surface spectra have been collected. We have also registered low-altitude, 2m resolution, AVIRSS (Airborne Visible/Infrared Imaging Spectrometer) spectra. We have investigated algorithms including the Spectral Angle function to compute the similarity between spectra. In this way we can correlate a surface measurement to its surroundings to determine novelty. In Figure 2, areas distinct from the sample spectra are highlighted and would indicate areas to seek novelty.

Addressing novelty detection, we developed and implemented methods of multi-class classification without knowledge of the number of classes comparing Iterative Gaussian Processes and Dirichlet Processes and a new adaptive technique. One application of this analysis is in determinng the most informative samples to collect (Figure 3).