Search

Hyperspectral imagery (HSI) has proven to be a useful tool when considering the task of target detection. Various processes have been developed that manipulate HSI data in different ways in order to render the data useable for target detection activities. A fundamental initial step in each of these processes is ensuring that the HSI data set obtained is in the same domain as the target’s spectral signature. In general, remotely sensed HSI is collected in terms of digital counts which are calibrated to units of radiance, whereas spectral target signatures are normally available in units of reflectance. This work investigates target detection using simulated hyperspectral imagery captured from highly oblique angles. Specifically, this thesis seeks to determine which domain, radiance or reflectance, is more appropriate for the off-nadir case. An oblique atmospheric compensation technique based on the empirical line method (ELM) is presented and used to compensate the simulated data used in this study. The resulting reflectance cubes are subjected to a variety of standard target detection processes. A forward modeling technique that is appropriate for use on oblique hyperspectral data is also presented. This forward modeling process allows for standard target detection techniques to be applied in the radiance domain. Results obtained from the radiance and reflectance domains are comparable. Under ideal circumstances, however, the radiance domain results observed tend to be superior compared to results observed in the reflectance domain. These somewhat favorable results observed in the radiance domain, considered with the practicality and potential operational applicability of the forward modeling technique presented, suggest that the radiance domain is an attractive option for oblique hyperspectral target detection.