The following seminars have been planned particularly oriented to both graduate and undergraduate students.
1- Wavelets and applications in signal and image processing
 "Wavelets and applications in signal and image processing",
Prof. Dr. A. Enis Çetin from Bilkent University. |
2- Statistical methods for error measuring and comparison in
classifiers
"Statistical methods for error measuring and comparison in
classifiers" Prof. Dr. Ethem Alpaydın from Bogazici University. |
3- Next Generation Wireless Sensor Networks
 "Wireless sensor Networks" Assoc.Prof.Dr. Özgür B. Akan from
Middle East Technical University |
4- Multihop Mesh Networks And Cooperative Communications
"The New Wireless Radio Access Network Paradigm: Multihop Mesh Networks And Cooperative Communications" Assoc.Prof.Dr. Halim Yanikomeroglu from Carleton University |
5- Empirical Mode Decomposition: an adaptive approach to analyze non-linear
time series
Dr. Paulo Gonçalves
Empirical Mode Decomposition is a recent technique (N. E. Huang et al.,
1998) introduced to analyze non-stationary and non-linear time series in a
totally adaptive way. In contrast to standard kernel based approaches (e.g.
wavelet decompositions), EMD is a fully data-driven method that recursively
decomposes a complex signal into a variable but finite number of zero-mean
with symmetric envelopes AM-FM components called Intrinsic Mode Functions
(IMF). [To proceed, an iterative algorithm locally identifies in the signal the
fastest oscillations and isolates them in the first IMF. Each successive IMF is
then obtained iterating the same sifting process on the remaining lower trend.]
This appealing analyzing tool is reversible by construction, and gives rise to
a
natural "scale" decomposition that goes beyond classic spectral analysis
and its
Fourier modes.
After a schematic presentation of the algorithm, we will address some of its
technical issues and report on the major EMD weakness, that is the lack of a
theoretical framework to support the method and to analytically characterize
an IMF.
To finish, we will present an EMD application to satellite time series imagery
for land cover classification. This simple study will not only illustrate the
flexibility of this non parametric method but also show, by comparison with
model-based identification procedures, the EMD ability at retrieving non-linear
modes. |
