| Summary: | Remote sensing of materials and chemical clouds using hyperspectral imaging sensors has many different applications. Some of these applications, such as detecting plumes in the aftermath of natural disasters, are time sensitive. As of now, the detection must be done on an offline system. In this thesis, we consider the matched filter, the normalized matched filter, and sequential maximum angle convex cone detection techniques for parallel implementations. The detectors are
mapped to a multicore CPU using multithreading and efficient data managements to achieve a real time result. A Graphics Processing Unit (GPU) is also investigated as a possible architecture for the time sensitive problem. Finally, we assess the performance of the implementations in terms of run time, and conclude how the performance can be improved further.
|
|---|
