"Angle gather reconstruction"

"This study constructs a diverse synthetic dataset for deep-learning-based pre-stack angle gather reconstruction based on the Zoeppritz equations and the convolutional model. Multiple practical factors, including noise contamination, angle-dependent wavelet variations, residual moveout, and data missing, are systematically incorporated during sample generation, allowing the dataset to represent the complex degradation patterns commonly observed in real prestack angle gathers. Based on this dataset, a U-net is further employed to verify its effectiveness for model training and generalization. Results from both synthetic and field-data experiments show that the constructed dataset effectively improves the reconstruction performance and generalization ability of the model. The trained model also improves the structural continuity, waveform consistency, and angle-dependent amplitude stability of prestack angle gathers, providing reliable input for subsequent prestack inversion and reservoir characterization."