On the evaluation of synthetic hyperspectral imagery

In developing algorithms that exploit model-generated data, it is important to understand the realism of the data generated by that model. One way to address this issue is to exercise a well understood, yet diverse process, that will help draw out the strengths and weaknesses of the data generation system. We accomplish this by using a typical chain of processing steps on a synthetic hyperspectral image created by the Digital Imaging Remote Sensing Image Generation (DIRSIG) tool [1]. The clustering, classification, and feature selection, which are part of this processing, are used to assess the realism of the data based on the performance compared to previous similar analysis on real hyperspectral data