Classification of acoustic signatures for the retrieval of environmental characteristics
Marielle Malfante, équipe SIGMAPHY / Gispa-Lab, Oct. 2015 - Oct. 2018
Phd student Marielle Malfante, équipe SIGMAPHY / Gispa-Lab, Oct. 2015 - Oct. 2018
Supervision : J.I. Mars, M. Dalla Mura, GIPSA-LAB, C. Gervaise Chaire Chorus.
Funding : 50% DGA, 50 % LabEx
Doctoral School : Terre Univers Environnement, Grenoble
The increasing number of civil and military operations in coastal environment requires a careful control of the SONAR emissions, due to their effects on the living conditions of marine mammals. Thus, this study aims at developing new methodologies that can ensure the protection of the underwater environment while preserving the effectiveness of the operations.
To this end, the DGA and, more in general, the scientific community is particularly interested in developing innovative methods of detection, classification and localization of marine mammals. Indeed, by adequate data processing it is possible to extract multiple information on the coastal zones (e.g., estimation of the parameters of the underwater environment) and on the animals populating those areas (e.g., monitoring animals’ behavior - migration, habitat occupation, ethology, analysis of the movements’ trajectories, environmental sonar).
A first study [Dadouchi et al 2012] has shown the capability to automatically recognize signatures of marine mammals with a good accuracy from acoustic signals. The method relies on the detection by statistical estimation of the signals time-frequency characteristics. Once detected, it is then possible to retrieve the characteristics waveforms by using two different approaches : i) time warping [Bonnel 2010] and ii) reconstruction time-frequency – phase [Ioana 2010].
Core of the thesis work
The core of this PhD is devoted to develop a classification method based on sparse representations (which is a booming approach in Machine Learning) [Barthelemy et al 2013a, b, c] for the discrimination of acoustic signals. One of the challenges in processing acoustic signals is to extract the useful information (e.g., the acoustic signature of the mammals). The informative components of the signals often live in subspaces of smaller dimensions (manifolds) and they can be represented parsimoniously, i.e., with little informative components selected from a set of components (i.e., a "dictionary") learned ad hoc. Such parsimonious representation can be subsequently exploited for classification [Song et al 2014].
[Barthelemy et al. 2013a] Barthelemy Q., Larue A. and Mars J.I., 2013, Decomposition and Dictionary Learning for 3D Trajectories, Signal Processing, Vol. 98, pp 423-437.
[Barthelemy 2013b], Barthelemy Q., 2013, Représentations parcimonieuses pour les signaux multivariés, Thèse de l’Université de Grenoble, Mai 2013.
[Barthelemy et al 2013c], Barthelemy Q., Gouy-Pailler C., Isaac Y., Souloumiac A., Larue A. and Mars J.I., 2013, Multivariate temporal dictionary learning for EEG, Journal of Neurosciences Methods, Vol. 215, Issue 1, pp 19-28.
[Barthelemy et al 2012], Barthelemy Q., Larue A., Mayoue A., Mercier D. and Mars J.I., 2012, Shift & 2D Rotation Invariant Sparse coding for multivariate signals, IEEE Trans. Signal Processing, 60 ( 2012), 4, 1597-1611.
[Bonnel 2010], Bonnel J., Analyse de la dispersion acoustique UBF (0-150 Hz) pour la surveillance et la caractérisation du milieu marin. Thèse de l’INP-Grenoble, Oct 2010.
[Dadouchi et al 2012], Dadouchi F., Gervaise C., Ioana C., Huillery J. and Mars J.I., 2013, Automated segmentation of linear time-frequency representation of marine mammal sounds, The Journal of the Acoustical Society of America, Vol. 134 (1), pp 77-87.
[Ioana et al 2010] C. Ioana, A. Jarrot, C. Gervaise, Y. Stéphan, A. Quinquis, "Localization in underwater dispersive channels using the time-frequency-phase continuity of signals". Ieee Trans. Signal Processing, 58 (2010) 4093-4107.
[Song et al 2014] B. SONG, J. LI, M. Dalla Mura, P. LI, A. Plaza, J. Bioucas-Dias, J. A. Benediktsson, and J. Chanussot, “Remotely sensed image classification using sparse representations of morphological attribute profiles,” Geoscience and Remote Sensing, IEEE Transactions on, accepted.
Sept 2015-Sept 2016
- Bibliography and submission of paper from the Master’s thesis.
- Tests with different classification algorithms (RF, SVM, NN, etc.)
- Dictionary learning approach for the context.
- System Architecture Modification to enhance the temporal Approach
- Go to the intra-classification.
- Field and Measurement campaign.
Sept 2016 - sept 2017
- From classification to the environmental knowledge
- Tools, test, validation
- Deep learning. Is it a solution ?
- International part (classifications & applications)
- Annual Report,
Sept 2015 - sept 2016
- Operational and “deliverables” part
- Thesis’s Report
- Conference and Papers.
- Defense of the Ph.D.