Digital Signal Processing Reference
In-Depth Information
When the basis set is well matched to the analyzed signal, the perfor-
mance of adaptive time-frequency algorithms can be very good. Compared
to the nonparametric methods, however, adaptive algorithms are usually
computationally quite expensive, especially if the number of bases in the
dictionary is large. A fast scheme is proposed in [5] for simple chirplet
functions to overcome this problem. Fast algorithms for more general basis
functions are still very much needed.
9.2
Back-Scattering Feature Extraction
Chapter 4 showed that time-frequency analysis is a very useful tool in
unveiling the underlying scattering phenomenology in radar back-scattering
data. This led to a better understanding of the various scattering mechanisms
in complex targets. Without time-frequency analysis, these mechanisms were
not easy to interpret from measured or computed data, nor could they be
easily studied from analytical solutions of Maxwell's equations due to target
complexity. This line of research should be continued to help build up a
more comprehensive knowledge base for complex shapes and exotic materials.
For target identification applications, work to incorporate the extracted time-
frequency features to improve the performance of existing classifiers is also
worthwhile pursuing [6, 7]. Finally, a more thorough understanding of the
scattering phenomenology from the electromagnetics point of view will allow
us to devise better basis functions in model-based time-frequency algorithms.
This can lead to physics-based signal processing algorithms that significantly
out-perform existing time-frequency tools.
Another potential application area where time-frequency analysis might
play a useful role is the suppression of clutter and propagation effects. For
instance, ocean surface often gives rise to large clutter that makes the detection
and classification of small floating targets a very difficult task [8]. In ground
penetrating radar for detecting buried objects, the dispersive effects due to
wave propagation through soil can lead to significant image distortion [9].
In foliage penetration (FOPEN) applications, the two-way propagation
through tree canopies will be an important factor on how well hidden targets
can be detected [10]. It would be fruitful to exploit the difference in target,
clutter, and propagation channel characteristics in the context of time-
frequency space to achieve clutter suppression, propagation effect removal,
and target feature enhancement. Some work along this line on SAR clutter
suppression using wavelets has been reported in [11].
Search WWH ::
Custom Search