Answers are grounded in this paper and cite their sources.
In seismic exploration, fracture characterisation remains challenging because of the complex morphologies and poorly constrained spatial distributions of fractures. Amplitude curvature attributes effectively reveal lateral heterogeneity and lithological variations critical for fracture interpretation. However, conventional methods for extracting amplitude curvature attributes rely heavily on additional information, such as geological horizons or dip angles, and require considerable interpreter expertise and time. In addition, the extracted curvature attributes may be affected by incoherent random noise, resulting in reduced spatial continuity and attribute quality. To overcome these challenges, we propose a deep learning method for amplitude curvature extraction based on a 3D Convolutional Neural Network (3D CNN). First, amplitude preprocessing based on principal component analysis (PCA) is used to enhance spatially coherent seismic reflection energy and suppress incoherent random noise, thereby improving the quality of the training labels. Second, a classification-based 3D CNN is developed to predict amplitude curvature attributes directly from post-stack 3D seismic data. The network is trained using amplitude curvature labels generated from seismic data after PCA-based amplitude preprocessing, enabling end-to-end amplitude curvature extraction. Field data examples demonstrate that the predicted amplitude curvature attributes are consistent with conventional results and enhance the delineation of fractures, minor faults, and channel features. After training, the proposed method reduces the amplitude curvature extraction time by approximately 76% for a 3D seismic volume of the same size, without requiring horizon or dip information during inference. These advantages facilitate the detailed characterisation of geological structures and improve the efficiency of seismic interpretation workflows.
No takes yet. Share an insight, caveat, or question.
Wang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: