Deep Learning for Enhancing Multisource Reverse Time Migration

Reverse time migration (RTM) is a technique used to obtain high-resolution images of underground reflectors; however, this method computationally intensive when dealing with large amounts seismic data. Multi-source RTM can significantly reduce the computational cost by processing multiple shots simultaneously. However, multi-source-based methods frequently result in crosstalk artifacts migrated images, causing serious interference imaging signals. Plane-wave migration, as mainstream multi-source method, yield plane waves different angles implementing phase encoding source and receiver wavefields; requires trade-off between efficiency quality. We propose based on deep learning for removing enhancing image quality plane-wave images. designed convolutional neural network that accepts an input seven at outputs clear enhanced image. built over 500 1024×256 velocity models, employed each them using produce raw 0°, ±10°, ±20°, ±30° network. Labels are computed from corresponding reflectivity models convolving Ricker wavelet. Random sub-images size 512×128 were training Numerical examples demonstrated effectiveness trained removal enhancement. The proposed superior both conventional (PWRTM) resolution. Moreover, only migrations, improving efficiency. In numerical examples, required our was approximately 1.6% 10% PWRTM, respectively.