Graduate Exam Abstract
December 20, 2013, 10:00 AM - 12:00 PM
ECE conference room
Characterization of Multiple Time-Varying Transient Sources from Multivariate Data
Abstract: This work considers the development of novel
methods for automatically detecting, classifying,
and estimating the signatures of highly variable
transient sources of intrinsic and extrinsic sound
within national park soundscapes. Monitoring
stations are typically deployed for months at a
time to constantly record acoustical events, where
historically (due to storage limitations) a
particular type of lossy time-transform
representation is used in order to avoid storing
raw audio waveforms. The goal is to prevent manual
observation and evaluation of a large volume of
data for detecting and classifying events produced
by commonly occurring man-made (e.g. aircraft) and
natural (e.g. weather effects) acoustical sources.
Automated source characterization is complicated
by many factors including nonstationary source
signatures, a large number of possible event types
(signal and interference), variable number of
sources that may be simultaneously present leading
to superimposed signatures, variable event
structure and duration even among those events
associated with a single source type, unknown
arrival times and Doppler shift, and the presence
of noise as well as other environmental and
operational variations. Since most applicable
methods typically only address a subset of these
complications, the objective is to develop
comprehensive solutions that consider all the
intricacies of natural acoustic scenes, while
still remaining viable for a multitude of
applications, e.g., speech recognition and
battlefield surveillance. <br><br>
Two primary solutions have been developed. In the
first method, a set of likelihood ratio tests is
hierarchically applied to each observation to
detect and classify both signal and interference
sources, that may be simultaneously present. Since
the signatures of each acoustical event typically
span several adjacent observations they are
assumed to be dependent, and hence, an adaptive
parameter estimation procedure is used to
calculate the likelihood values. The second method
evaluates the likelihood of observing a sequence
of elementary atom coefficient states, that are
extracted from the original vector observation
sequence, given a parameter set for each signal
type. This framework remains robust to
interference, and indeed is capable of separating
signal and interference signatures, by exploiting
sparsity that is inherently present in the data.
The experimental results of applying each method
to the problem of characterizing national park
soundscapes attest to their effectiveness at
performing the mentioned tasks.
Adviser: Dr. Mahmood Azimi
Non-ECE Member: Dr. F. Jay Breidt, Statistics
Member 3: Dr. Ali Pezeshki, ECE
Addional Members: Dr. Kurt Kristrup, ECE
1. N. Wachowski and M.R. Azimi-Sadjadi, "Detection and classification of nonstationary transient signals using sparse approximations and Bayesian networks," IEEE Trans. Audio, Speech, and Lang. Process., to be summited, December 2013.
2. N. Wachowski and M. Azimi-Sadjadi, "Characterization of multiple transient acoustical
sources from time-transform representations," IEEE Trans. Audio, Speech, and Lang. Process., vol. 21, no. 9, pp. 1966-1978, September 2013.
3. N. Wachowski and M.R. Azimi-Sadjadi, "A new synthetic aperture sonar processing method using coherence analysis," IEEE Journal of Oceanic Engr., vol. 36, no. 4, pp. 665-678, October 2011.
4. J. Cartmill, N. Wachowski, and M. R. Azimi-Sadjadi, "Buried underwater object classification using a collaborative multiaspect classifier," IEEE Journal of Oceanic Engr., vol. 34, no. 1, pp. 32-44, January 2009.
5. Y. Zhao, A. Dinstel, M.R. Azimi-Sadjadi, and N. Wachowski, "Localization of near-field sources in sonar data using the sparse representation framework," Proc. of MTS/IEEE Oceans, pp. 1-6, September 2011.
6. N. Wachowski, M.R. Azimi-Sadjadi, and R. Holtzapple, "SAS-like acoustic color processing for a single-hydrophone sonar platform," Proc. of MTS/IEEE Oceans, pp. 1-8, September 2010.
7. M.R. Azimi-Sadjadi and N. Wachowski, "An information theoretic approach for in-situ underwater target classification," Proc. of the International Joint Conference on Neural Networks (IJCNN), pp. 1-8, July 2010.
8. Y. Zhao, N. Wachowski, and M.R. Azimi-Sadjadi, "Target coherence analysis using canonical correlation decomposition for SAS data," Proc. of MTS/IEEE Oceans, pp. 1-7, October 2009.
9. Y. Zhao, M.R. Azimi-Sadjadi, N. Wachowski, and N. Klausner, "Spatial correlation analysis using canonical correlation decomposition for sparse sonar array processing," IEEE International Conference on Systems, Man, and Cybernetics, pp. 2739-2744, October 2009.
10. N. Wachowski, and M.R. Azimi-Sadjadi, "A likelihood-based decision feedback system for multi-aspect classification of underwater targets," Proc. of the IJCNN, pp. 3232-3239, June 2009.
11. N. Wachowski, and M.R. Azimi-Sadjadi, "Buried underwater object classification using frequency subband coherence analysis," Proc. of MTS/IEEE Oceans, pp. 1-8, September 2008.
12. J. Cartmill, N. Wachowski, and M.R. Azimi-Sadjadi, "Buried underwater object classification using a collaborative multi-aspect classifier," Proc. of the IJCNN, pp. 1807-1812, August 2007.
13. N. Wachowski, J. Cartmill, and M.R. Azimi-Sadjadi, "Underwater target classification using the wing BOSS and multi-channel decision fusion," Proc. Of SPIE Defense and Security, Vol. 6553, pp. 65530Q.1-65530Q.10, April 2007.
Program of Study: