In the previous paper [1], deep learning was applied to predict the propagation of atmospheric acoustic waves. A deep learning network was trained on ground-truth composed of a massive set of wave propagation simulations performed using a Parabolic Equation code. Once the ML model is established and trained, a very rapid estimate of the amplitude of the propagated acoustic wave can then be available. This is useful because, for instance, knowing the ground pressure levels associated with earthquakes, man-made or volcanic explosion properties, and ocean-generated microbarom wavefields. However, the computational cost inherent in full-waveform modelling tools often prevents the exploration of a large parameter space. Therefore, having a reliable machine-learning based prediction can be of great value.
In this master project, you will be invited to
- assess the applicability of these methods to other domains of acoustic wave propagation, for example, in medical ultrasound imaging,
- explore the impact of existing deep learning architectural choices to predicting the spatial distribution of infrasound attenuation,
- design new architectures to improve the predicted model,
- create new data through executing the existing modelling tools to evaluate and train the models,
- investigate whether full green's function prediction is feasible, and
- look into machine-learning based approaches to speed up the simulation of of non-linear wave propagation.
The scope of the changes and types of models or components to evaluate will be discussed and detailed with the candidates, and will depend on their background.
Qualifications:
- Excellent Python, and tensorflow or pytorch knowledge
- Signal processing (desired)
- Image processing (desired)
- Wave propagation understanding (desired)
- Git
Literature:
[1] Quentin Brissaud, Sven Peter Näsholm, Antoine Turquet, Alexis Le Pichon, Predicting infrasound transmission loss using deep learning, Geophysical Journal International, 2022;, ggac307, https://doi.org/10.1093/gji/ggac307