Automatic semblance velocity analysis using Convolutional Neural Networks

Abstract

Velocity analysis can be a time-consuming task when it is performed manually. Methods have been proposed to automate the process of velocity analysis, which, however, typically requires significant manual effort. We propose using the Convolutional Neural Network (CNN) to estimate stacking velocities directly from the semblance. Our CNN model uses two images as one input data for training. One is the entire semblance (guide image), and the other is a small patch (target image) extracted from the semblance at a specific time step. Labels for each input dataset are the root mean square (RMS) velocities. We generate the training dataset using synthetic data. After training the CNN model with synthetic data, we test the trained model with other synthetic data that was not used in the training step. The result shows that the model can predict a consistent velocity model. One also notices that when the input data is extremely different from the one used for the training, the CNN model will hardly pick the correct velocities. In this case, I propose to adopt transfer learning to update the trained model (base model) with a small portion of the target data. The latter improves the accuracy of the predicted velocity model. The Marmousi dataset and a marine dataset from the Gulf of Mexico are used for validating the proposed automatic velocity analysis algorithm.