Automatic fault detection in seismic images using deep learning-based methods has attracted great interest in recent years. Detecting faults by deep learning methods is generally considered a supervised learning task, which requires numerous diverse training samples and corresponding accurate labels. Because field data with faults labeled by experienced interpreters are difficult to acquire, training samples are usually generated synthetically, in which the 1D seismic wavelets are convolved with reflectivity models. However, the 1D-convolution-based synthetic seismic images still differ from the real seismic images. Therefore, the network trained by such samples may fail to detect some faults in the field data. Full-wavefield approaches are the optimal seismic modeling methods, but they cannot be widely used at present due to the high cost. In this paper, we present an efficient approach to generate realistic synthetic seismic images by using a point-spread function (PSF)-based convolution. Because the size of a seismic sample used for deep learning training is a small range relative to the field seismic data, assuming the background velocity is close to homogeneous in such a small range, we can efficiently construct the PSF using the analytical Green’s function. With an Intel Core i9-10900K CPU, the PSF approach takes approximately 15 min to generate a seismic image of size 128 × 128 × 128, whereas it takes approximately 20 h to generate a seismic image of the same size using reverse time migration. The examples of one synthetic image and two field seismic images demonstrate that the network trained with the PSF convolution samples can predict more accurate and continuous faults than that trained with the 1D convolution samples.

Deep-learning-based seismic data interpretation has received extensive attention and focus in recent years. Research has shown that training data play a key role in the process of intelligent seismic interpretation. At present, the main methods used to obtain training data are synthesizing seismic data and manually labeling the real data. However, synthetic data have certain feature differences from real data, and the manual labeling of data is time-consuming and subjective. These factors limit the application of deep learning algorithms in seismic data interpretation. To obtain realistic seismic training data, we propose label-to-data networks based on cycle-consistent adversarial networks in this work. These networks take random labels and unlabeled real seismic data as input and generate synthetic seismic data that match the random labels and have similar features to the real seismic data. Quantitative analysis of the generated data demonstrate the effectiveness of the proposed methods. Meanwhile, test results on different data indicate that the generated data are reliable and can be applied for seismic fault detection.

Intracratonic strike-slip faults play a vital role in hydrocarbon migration and accumulation, making their accurate detection crucial for subsurface structure interpretation and reservoir characterization. Although numerous approaches have been developed for automatic fault interpretation, they are still unable to fully recognize intracratonic strike-slip faults with subtle reflection variations in seismic images. We develop a supervised convolutional neural network (CNN) to automatically and precisely delineate these faults in 3D seismic images. The primary obstacle in using a CNN for this task is the lack of diverse seismic training samples and corresponding high-quality labels. To address this, we develop an efficient two-step workflow for automatically generating a tremendous amount of 3D seismic training sample pairs: (1) guided by the prior geologic structure patterns of the study area, we use parameterized vector fields to simulate various velocity models with accurate fault labels incorporating realistic intracratonic strike-slip faulting features, such as the transtensional, transpressional, and strike-slip translational types and (2) following seismic imaging laws, we synthesize seismic training samples from simulated velocity models through the point-spread function-based convolution method. One synthetic and two field examples demonstrate that the CNN trained with this seismic data set containing prior geologic patterns and seismic imaging features outperforms the conventional fault detection approaches in accurately and continuously delineating intracratonic strike-slip faults.

Among deep-learning-based seismic fault detection approaches, most of the training datasets are generated by using synthetic methods in order to enhance the diversity of training data. However, due to the feature differences between synthetic seismic data and real data, and the difficulty of synthesizing all types of actual geological structures, networks trained using synthetic data exhibit poor generalization performance in practical data applications. In this study, we propose a barely-supervised-learning-based seismic fault detection scheme. The training data are extracted from the real seismic data and the fault labels are manually annotated by interpreter. In this scheme, only 10% of the training data need to be labeled. Furthermore, to avoid dense annotation of faults on 3D seismic data, we adopt an orthogonal annotation strategy, where only one inline section and one horizontal slice are annotated for one 128×128×128 training sample. As a result, the pixels that have been actually annotated constitute merely 0.1563% of the total voxels within the training sample. Then the orthogonal annotation is extended to the dense annotation of training data by using a fault registration method. We confirmed the feasibility and effectiveness of this approach using synthetic data. Accurate fault detection can be achieved by annotating a few orthogonal sections and slices from actual samples. Further application on two real seismic data demonstrated that the fault detection by this method is more continuous and accurate compared to the results predicted by other methods.