Application of Deep Learning in First-Break Picking of Shallow OBN Data

As well known, the modeling of the near-surface from first-break plays a significant role on the sub-surface imaging, reservoir characterization, and monitoring. Small errors in first-break picking can greatly impact the seismic velocity model building, so it is necessary to pick high-quality travel times. Geoscientists from around the world continues trying their best to address the near-surface challenges. Due to the rapid development of high-efficiency acquisition technique, such as WBH (wide-azimuth, broadband and high-density) acquisition technique and blended source acquisition technique, the quantity of seismic data, especially 3D seismic exploration, has leapt from GB to TB(some to PB), which sets a big challenge for first-break picking. Traditional first-break picking methods can't meet the production. In recent years, with the development of computer capacity and algorithm, artificial intelligence has changed our lives in many ways. In seismic exploration, artificial intelligence, like deep learning, has played a more and more important role now, from fault prediction, attribute identification to velocity and first break picking. Generally, deep learning is a new neural network which has multiple hidden layers, mostly over 3 layers, compared with traditional neural network. Deep learning includes Deep Belief Network (DBN), Convolutional Neural Network (CNN), Recurrent Neural Network (RNN) and so on. In this paper, Convolutional Neural Network (CNN) and Recurrent Neural Network (RNN) have been combined for first-break picking in a large 3D OBN project of Caspian Sea. A high precision near seabed velocity model is built based on the auto-picked first break with tomography inversion, which provides a good solution for static problem of the survey.