Prediction of sedimentary facies based on GAF-ResNet
Ziyi Li, Jiachun You, Haipeng Hu, Yiyang Jiang, Yulin Wu, Shanzheng Hu, Xiangwen Li · Journal of Geophysics and Engineering · 2025
Abstract Sedimentary facies classification is of great significance in hydrocarbon exploration, seismic surveying, and geological research. Accurate facies classification helps unveil the depositional environment and reservoir distribution of subsurface strata, thereby enhancing the precision and efficiency of resource exploration. To overcome the limitations of attribute-driven methods and shallow machine learning models in complex geological settings, this study proposes a novel deep learning-based approach, GAF-ResNet, which integrates Gramian angular field (GAF) transformation with a residual neural network (ResNet). Raw seismic traces are encoded into two-dimensional GAF feature maps, which are then fed into a ResNet to exploit their time-frequency features for improved classification performance. Comparative experiments with five models—transformer, 1D ResNet, support vector machine, long short-term memory, and self-organizing map—demonstrate the superiority of GAF-ResNet in classification accuracy, generalization capability, and boundary resolution. These advantages are further validated through t-SNE visualization and clustering metrics. The pretrained model is subsequently applied to area-wide sedimentary facies prediction. This research offers an effective solution for high-precision subsurface characterization and provides a robust foundation for subsequent geological interpretation.