A comparative study among EMDs for micro-doppler UAV classification
Due to the mass production and unregulated possession of mini unmanned aerial vehicles (UAVs). These aerial vehicles could be a threat when abused, so to detect and classify them has become important. To build an automatic UAV classification system, researchers have developed many methods for mic...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Theses and Dissertations |
Language: | English |
Published: |
2017
|
Subjects: | |
Online Access: | http://hdl.handle.net/10356/72616 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Due to the mass production and unregulated possession of mini unmanned aerial vehicles
(UAVs). These aerial vehicles could be a threat when abused, so to detect and
classify them has become important. To build an automatic UAV classification system,
researchers have developed many methods for micro-Doppler analysis using spectrogram,
spectrogram, wavelet and so on. Among them, spectrogram analysis is one of the most
frequently selected method for its efficiency and performance. Spectrogram could be obtained
by performing short-time Fourier transform (STFT). Besides basic Fourier analysis
method, Empirical Mode Decomposition (EMD) is another time-frequency (TF) algorithm.
Different from other methods, EMD is an adaptive method which means no need
for background information of input data. This algorithm has been applied to many different
areas, for instance, biomedical and image processing. However, this algorithm is
not fully tested and developed under radar micro-Doppler analysis.
Aside from the original EMD algorithm, there are many extensions. These methods, for
instance, ensemble EMD (EEMD) and bivariate EMD (BEMD) are proposed for different
applications and have their own advantages and drawbacks. Since these extensions
are application dependent and even original EMD is new for UAV classification using
micro-Doppler signature, all these algorithm could be quite potential. So it is necessary
to test all these methods and compare them in terms of accuracy and computational complexity.
BEMD is the complex-valued extension, so some features extracted directly from
complex-valued sequences is also required.
Some observations could be made from the result we obtained, when the signal to noise
ratio (SNR) is low and data length is sufficient, noise-assisted method EEMD become a
good choice. For most cases with complex-valued input, BEMD provides better performance.
Besides the comparison, some features are extracted from complex-valued input
in using circularity property of complex-valued variables. |
---|