Transformer assisted dual U-net for seismic fault detection

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