Missile defense applications require flash radiometry for target kill assessment and spectral analysis of high impact scenarios. The goal is to collect a set of registered spectrally contiguous images of a scene's spatial and radiation distribution within the shortest possible time. In contrast with classical broadband imaging, the spectral information permits the classification of the scene based on material content or temperature. Hence, the target data to be acquired by an imaging spectrometer form a three dimensional (x, y, and l) object referred to as the object cube.
The Computed Topography Imaging Spectrometer (CTIS) is one innovative sensor built by the University of Arizona (UOA) that is being considered by the Missile Defense Agency for such applications. It employs a simple concept (diffractive optics through a computer generated hologram) to project the object cube on the sensor's focal plane for measurements that involve no scanning functions, and therefore introduce no time-related artificial distortions to the phenomenon being detected. The CTIS is a snapshot (flash) instrument and its potential applications are the investigation of spatially complex dynamic events. These events contain thermal or chemical structure that are classified by spectral radiance. The way that CTIS measures target objects can be time consuming and requires complex post-processing because objects are not measured directly. Consequently, there is a strong rationale to accelerate existing algorithms and tailor them to novel computational hardware that would enable real-time performance.
The most recent reconstruction approach (developed by UOA and others) uses one of two stasticial algorithms: mixed expectation maximization (MEM) and multiplicative algebraic reconstruction technique (MART). The former corresponding software labeled MERT (mixed expectation reconstruction technique) is written in a combination of IDL, C, and C++. In results reported to date, the typical timing per reconstructed image ranges from 50 s for a very small (toy) target object (46 x 46 x 21 voxels, 512 x 512 focal plane) using 10 MERT iterations.