With the dramatic growth and complexity of seismic data, manual seismic facies analysis has become a significant challenge. Machine learning and deep learning (DL) models have been widely adopted to assist geophysical interpretations in recent years. Although acceptable results can be obtained, the uninterpretable nature of DL (which also has a nickname ``alchemy'') does not improve the geological or geophysical understandings on the relationships between the observations and background sciences. This article proposes a noble interpretable DL model based on 3-D (spatial-spectral) attention maps of seismic facies features. Besides regular data-augmentation techniques, the high-resolution spectral analysis technique is employed to generate multispectral seismic inputs. We propose a trainable soft attention mechanism-based deep dilated convolutional neural network (ADDCNN) to improve the automatic seismic facies analysis. Furthermore, the dilated convolution operation in the ADDCNN generates accurate and high-resolution results in an efficient way. With the attention mechanism, not only the facies-segmentation accuracy is improved but also the subtle relations between the geological depositions and the seismic spectral responses are revealed by the spatial-spectral attention maps. Experiments are conducted, where all major metrics, such as classification accuracy, computational efficiency, and optimization performance, are improved while the model complexity is reduced.

As a powerfully diagnostic tool for structural interpretation, reservoir characterization, and hydrocarbon detection, quality factor Q provides useful

The synchrosqueezing generalized S-transform (SSGST) is commonly used to generate an isofrequency component of a signal by squeezing the decomposed frequency components of the signal. However, for seismic signals, the single-trace process can have a lack of lateral information in the squeezed results and lead to some discontinuous geologic information that will mislead the interpreter. Thus, to improve the stability of SSGST, we have developed a multichannel seismic trace squeezing method. Multichannel SSGST (MSSGST) considers the decomposed frequency components of neighboring traces of the analysis seismic trace and then reconstructs the center trace. Therefore, compared with SSGST, the multichannel processing improves the stability of the squeezing and produces more laterally continuous results that properly follow the geologic phenomenon. The effectiveness of MSSGST is validated using various field data. We use the application to demonstrate the potential of multichannel squeezing to perform well at (1) improving the energy focusing and continuity of the decomposed frequency components, (2) depicting the boundaries of geologic structures, and (3) identifying the thin layers.

Blending of different frequency components of seismic traces is a common way to estimate the relative time thickness of the formation. Red, blue, and green (RGB) color blending is one of the most popular blending models in analyzing multiple seismic attributes. Geologists and geophysicist interpreters typically associate low-frequency components (formations with the largest thickness value) with a red color, medium-frequency components (formations with a medium thickness value) with a green color, and high-frequency components (formations with the smallest thickness value) with a blue color for the thickness estimation of thin beds using frequency components. However, we found that the same result of RGB blending may come from different sets of three frequency components. As a result, the same blended color may correspond to several different time thicknesses. It is also very difficult to interpret the corresponding thickness of the blended colors such as white and yellow. To avoid the ambiguity of time-thickness estimation using RGB blending, we have estimated the time thickness of the thin beds using all of the frequency components in a user-defined frequency band instead of only three frequency components. Our workflow begins with the normal seismic spectral decomposition. Considering that the different reflectivity pairs with a different time thickness have a different amplitude spectrum, we then use the self-organizing map to cluster the decomposed amplitude spectra of seismic traces. We finally assign each cluster with a relative thickness by comparing the clustered results with well logs.