In our work, the deep learning technique has been used to develop an automatic method for identifying the first arrival times of seismic waves. This method introduces transfer learning to train a deep neural network, given a limited number of continuous wavelet transform (CWT) feature images as input. The application of the CWT for feature extraction, aimed at detecting abrupt changes in the amplitude, phase, and frequency produced by first arrivals as a whole rather than any single characteristic, provides the most informative images. First, we apply the CWT to each seismic trace to obtain the CWT feature images and split them into a set of subimages. Then, a pretrained convolutional neural network (CNN) is fine-tuned with limited labeled subimages. The resulting model can be used to predict probability distributions of noise, first-break, and post first-break. Finally, the first arrival times are extracted from the peaks of the probability distributions. We have tested the performance of the method using vibroseis, dynamite, and air gun shot records, which include various types of seismic waves and noise. More accurate and robust results can be obtained with the proposed method compared with the short-time and long-time average (STA/LTA) algorithm and the adaptive multiband picking algorithm (AMPA).

Automatic seismic fault identification for seismic data is essential for oil and gas resource exploration. The traditional manual method cannot accommodate the needs of processing massive seismic data. With the development of artificial intelligence technology, deep learning techniques based on pattern recognition have become a popular research area for seismic fault identification. Despite the progress made with U-shaped neural networks (Unet), they still fall short in meeting the stringent requirements of fault prediction in complex structures. We propose a novel approach by combining a standard Unet with a transformer Unet to create a parallel dual Unet model, called Dual Unet with Transformer. To improve the accuracy of fault prediction, we compare six loss functions (including Binary Cross Entropy loss, Dice coefficient loss, Tversky loss, Local Tversky loss, Multi-scale Structural Similarity and Intersection over Union loss) using synthetic data, based on three evolution metrics involving Dice coefficient, Sensitivity and Specificity, find that the binary cross entropy loss function is the most robust one. An example comparing the prediction performance of different Unet models on synthetic data demonstrates the superior performance of our Dual Unet model, verifying the practical application value. To further validate the practical feasibility of our proposed method, we use real seismic data with a complex fault system and find that our proposed model is more accurate in predicting the fault system compared to well-developed Unet models such as the classical Unet and classical coherence cube algorithm, without transfer learning. This confirms the potential for wide-scale application of our proposed model.

Sparse impedance inversion is widely applied for hydrocarbon prediction. Due to the low sparseness of traditional sparse constraints, pseudolayer and low-resolution problems still exist. To overcome this barrier, an impedance inversion method based on the reweighted L1-norm sparse constraint is proposed. Different from the traditional L1-norm constraint, which considers the location information of impedance boundaries, the reweighted L1-norm uses the amplitude information of impedance boundaries. The amplitude information can improve the sparseness, which helps inversion obtain more precise boundaries of impedance and weaken the pseudolayer phenomenon. Both the reweighted L1-norm constraint and initial model constraint are used to construct the objective function. The alternating direction method of multipliers (ADMM) algorithm is utilized to obtain the inversion algorithm. Both synthetic and field data tests prove that the impedance boundary of the proposed method is more accurate, and the pseudolayer phenomenon is weakened.

Seismic facies analysis is a crucial foundation for basin-fill studies and oil and gas exploration. With its rapid development, CNN-assisted interpretation is becoming increasingly popular. However, CNN models are often considered "black boxes" that lack transparency. To understand how CNN models classify seismic facies and visualize the contribution of each seismic attribute to the final predictive scoring, we have investigated class activation map (CAM) techniques and an explainable tool called Shapley additive explanations (SHAP) value. Based on real seismic data collected in the Sichuan basin, we compared the visualization performances of CAM and SHAP methods and found that the SHAP tool has better visualization capabilities than CAM methods, which only produce heat maps with positive values. Using SHAP values, we identified the importance of each seismic attribute and refined redundant attributes. This approach establishes a connection between seismic attributes and sedimentary environments and is a prime example of the capability of deep learning to discover knowledge beyond human experience. We applied the selected seismic attributes to generate a refined CNN model and compared it to the original CNN model, demonstrating the superiority of our proposed strategy. When we compared the predicted seismic facies using the refined CNN model based on SHAP features, the conventional K-means, SVM and Gaussian Naive Bayes methods, it is observed that our predicted map aligns well with geological knowledge with less prediction errors, demonstrating the effectiveness and feasibility of our developed strategy.

To investigate wavefield depth extrapolation using the full-wave equation, we have derived a new depth extrapolation scheme for migration using functions of the vertical wavenumber. We develop a complete matrix multiplication formulation and approach to calculate the related mathematical functions of the vertical wavenumber and perform depth extrapolation using matrix multiplication only. Because our depth extrapolation algorithm involves only matrix multiplication, it is naturally applicable to parallel computations. Impulse response experiments demonstrate that our proposed migration method can achieve the same accuracy as full-wave-equation migration using the finite-difference method, in terms of phase information, even for media with strong lateral velocity changes. In numerical experiments using a smoothed version of the 2D SEG/EAGE salt model, our migration method provides an equivalent imaging result compared with reverse time migration (RTM) and a more accurate imaging result than migration using one-way propagators. Our method has certain potential advantages over RTM using the same full-wave equation with fewer internal multiple scatterings and fewer data storage requirements. Our adopted method is a stable depth extrapolation scheme because the evanescent waves are well suppressed. The numerical experimental results on the synthetic model demonstrate the importance of suppressing evanescent waves in a full-wave-equation-based depth extrapolation scheme and migration for imaging quality and computation cost.

Because swell noises are very common in marine seismic data, it is extremely important to attenuate them to improve the signal-to-noise ratio (S/N). Compared to process noises in the time domain, we have built a frequency-domain convolutional neural network (CNN) based on the short-time Fourier transform to address swell noises. In the numerical experiments, we quantitatively evaluate the denoising performances of the time- and frequency-domain CNNs, compare the impacts of network structures on attenuating swell noises, and study how network parameter choices impact the quality of the denoised signal based on peak S/N, structural similarity, and root-mean-square-error indices. These results help us to build an optimal CNN model. Furthermore, to illustrate the superiority of our proposed method, we compare the conventional and proposed CNN methods. To address the generalization capability of CNN, we adopt transfer learning by using fine tuning to adjust the weights of the pretrained model with a small amount of target data. The application of transfer learning improves the quality of the denoised images, which further proves that our proposed method with transfer learning has the potential to be deployed in actual seismic data acquisition.