Semi-Supervised Salt Segmentation Using Mean Teacher

Detecting subsurface salt structures from seismic images is important for seismic structural analysis and subsurface modeling. Recently, deep learning has been successfully applied in solving salt segmentation problems. However, most of the studies focus on supervised salt segmentation and require numerous accurately labeled data, which is usually laborious and time-consuming to collect, especially for the geophysics community. In this paper, we propose a semi-supervised framework for salt segmentation, which requires only a small amount of labeled data. In our method, adopting the mean teacher method, we train two models sharing the same network architecture. The student model is optimized using a combination of supervised loss and unsupervised consistency loss, whereas the teacher model is the exponential moving average (EMA) of the student model. We introduce the unsupervised consistency loss to better extract information from unlabeled data by constraining the network to give consistent predictions for the input data and its perturbed version. We train and test our novel semi-supervised method on both synthetic and real datasets. Results demonstrate that our proposed semi-supervised salt segmentation method outperforms the supervised baseline when there is a lack of labeled training data.

Seismic Quantitative Interpretation for Uncertainty Reduction of Subsurface Modeling of the Deep Visean Reservoirs

The seismic data have historically been utilized to perform structural interpretation of the geological subsurface. Modern approaches of Quantitative Interpretation are intended to extract geologically valuable information from the seismic data. This work demonstrates how rock physics enables optimal prediction of reservoir properties from seismic derived attributes. Using a seismic-driven approach with incorporated prior geological knowledge into a probabilistic subsurface model allowed capturing uncertainty and quantifying the risk for targeting new wells in the unexplored areas. Elastic properties estimated from the acquired seismic data are influenced by the depositional environment, fluid content, and local geological trends. By applying the rock physics model, we were able to predict the elastic properties of a potential lithology away from the well control points in the subsurface whether or not it has been penetrated. Seismic amplitude variation with incident angle (AVO) and azimuth (AVAZ) jointly with rock-derived petrophysical interpretations were used for stochastical modeling to capture the reservoir distribution over the deep Visean formation. The seismic inversion was calibrated by available well log data and by traditional structural interpretation. Seismic elastic inversion results in a deep Lower Carboniferous target in the central part of the DDB are described. The fluid has minimal effect on the density and Vp. Well logs with cross-dipole acoustics are used together with wide-azimuth seismic data, processed with amplitude control. It is determined that seismic anisotropy increases in carbonate deposits. The result covers a set of lithoclasses and related probabilities: clay minerals, tight sandstones, porous sandstones, and carbonates. We analyzed the influence of maximum angles determination for elastic inversion that varied from 32.5 to 38.5 degrees. The greatest influence of the far angles selection is on the density. AI does not change significantly. Probably the 38,5 degrees provides a superior response above the carbonates. It does not seem to damage the overall AVA behavior, which result in a good density outcome, as higher angles of incidence are included. It gives a better tie to the wells for the high density layers over the interval of interest. Sand probability cube must always considered in the interpretation of the lithological classification that in many cases may be misleading (i.e. when sand and shale probabilities are very close to each other, because of small changes in elastic parameters). The authors provide an integrated holistic approach for quantitative interpretation, subsurface modeling, uncertainty evaluation, and characterization of reservoir distribution using pre-existing well logs and recently acquired seismic data. This paper underpins the previous efforts and encourages the work yet to be fulfilled on this subject. We will describe how quantitative interpretation was used for describing the reservoir, highlight values and uncertainties, and point a way forward for further improvement of the process for effective subsurface modeling.