|
|
|
Multiplierless Filters for Real-Time Processing of Hyperspectral Images
|
|
|
| There is a recent thrust by NASA to develop high-resolution hyperspectral imaging systems for use in geostationary vehicles. The concept of hyperspectral imagery is that the imaging interferometer has a sensor module which consists of a NxN array of sensors. Each of these sensors provides a Mx1 spectrum. Therefore, for each spectral component, there is an image representing the amplitude over
a mapped space. This represents a 2-D increase in information over conventional spectrometers which have only one sensor. However, the down side is that there is an incredible amount of data to process, compress, deliver, and analyze. Real-time processing of such data sets is nearly impossible with current signal processing algorithms and hardware technology.
Conventional algorithms use a variety of different filters for processing image data. These filters are implemented using many multipliers and adders. The multipliers are a bottleneck to high speed computations. A solution is therefore to design filters with coefficients that are powers-of-two. The multiplications can now be implemented using simple shifting operations. In VLSI mplementations,
multiplierless filters are ideal because of their high speed and low power. The design of 2-D multiplierless filters is a long standing open problem.
The main goal of this project is to develop the tools and techniques for designing 2-D multiplierless digital filters. These filters will then be used to process hyperspectral image data, both in the spatial
and temporal directions. The final goal is to derive algorithms using these filters for the restoration and compression of hyperspectral images.
Although multiplierless filters and algorithms are ideally suited for processing hyperspectral images from FT spectrometers, they will also be useful in other industrial applications that use, process, store
and transmit images and video. |
|
|
