Recent applications of machine learning algorithms in the seismic domain have shown great potential in different areas such as seismic inversion and interpretation. However, such algorithms rarely enforce geophysical constraints — the lack of which might lead to undesirable results. To overcome this issue, we have developed a semisupervised sequence modeling framework based on recurrent neural networks for elastic impedance inversion from multi-angle seismic data. 

In this research project, the physics that governs the data is fused with the deep neural networks to allow the networks to learn from a few measurement in a semi supervised scheme. 

Texture representation plays a very important role in many applications of the field of visual understanding. Examples of these applications are segmentation, retrieval, and material recognition. 

In this research project, we developed and designed discriminative, robust, and efficient texture features that can be used to build interpretable texture models. In addition, we design similarity measures designed specifically to capture textural content in an image as opposed to generic distance (or similarity) measures. 

Seismic interpretation, the process of identifying the different subsurface structures. The process is very time consuming and labor intensive.

In this project, we aim to use the fast growing field of deep learning to reduce the time and effort spend on seismic interpretation. We develop automatic seismic interpretation system. In addition, we create large-scale labeled datasets suitable for training deep learning models.  

Reservoir characterization involves the estimation petrophysical properties from well logs, core data and seismic data. Estimating such properties is a challenging task due to the non-linearity and heterogeneity of the subsurface. Recent advances in machine learning have shown promising results for recurrent neural networks (RNN) in modeling complex sequential data such as videos and speech signals.

In this research, we model seismic traces as sequential data and applying state-of-the-art sequence modeling techniques such RNNs and LSTMs for reservoir characterization including, but not limited no, property prediction and facies analysis.

Seismic attributes play a pivotal role in the field of seismic interpretation. Some of these attributes are derived from the seismic signals (traces) directly which makes them vulnerable to noise. 

In this project, we utilize multiscale analysis and other as well as other image processing representations to enhance the quality and resolution of these attributes without compromising small-scale features that are of interest in seismic images.