Fault Detection via 2.5D Transformer U-Net with Seismic Data Pre-Processing

Seismic fault structures are important for the detection and exploitation of hydrocarbon resources. Due to their development popularity in geophysical community, deep-learning-based methods have been proposed achieved SOTA results. efficiency benefits full spatial information extraction, 3D convolutional neural networks (CNNs) used widely directly detect faults on seismic data volumes. However, using training requires expensive computational resources can be limited by hardware facilities. Although 2D CNN less computationally intensive, they lead loss correlation between slices. To mitigate aforementioned problems, we propose predict a section multiple neighboring profiles, that is, 2.5D detection. In CNNs, convolution layers mainly extract local pooling may disrupt edge features data, which tend cause discontinuities. this end, incorporate Transformer module U-net feature extraction enhance prediction continuity. reduce discrepancies synthetic different real datasets, apply standardization workflow improve stability datasets. Netherlands F3 tests show that, when labels, U-net-based method predicts more subtle with higher continuity than baseline model.