Post-processing of migrated common image gathers is an essential step before residual moveout picking for pore pressure analysis and velocity modeling. Conventional processing workflow relies on time-intensive manual optimization of processing parameters.
To reduce the turnaround time, 2D and 3D neural networks (2D and 3D GAP) were developed to clean seismic gathers along offset and time/depth directions. However, the issues of low-amplitude and steep-dip washout persist and make the result less reliable. To mitigate these issues, we use AVC (amplitude-based scaling) and gather stack as additional features to balance the amplitude and signal-to-noise ratio and use 4D convolutional neural networks to leverage the information from adjacent inline and crossline gathers in addition to time/depth-offset domain.
The new model is trained and tested on six surveys and the test results show that the new model is generalizable to unseen test surveys and the washout issues are greatly mitigated. The new model greatly decreases the turnaround time and provides reliable automatic migrated gather processing results, which can be confidently used for velocity modeling.
