Using a synthetic data trained convolutional neural network for predicting subresolution thin layers from seismic data

Abstract

Numerous studies have demonstrated the capability of supervised deep learning techniques for predicting geologic features of interest from seismic sections, including features that are difficult to identify using traditional interpretation methods. However, the successful application of these techniques in practice has been limited by the difficulty of obtaining a large training data set where the seismic data and corresponding ground truth labels are well-defined. Manually creating large amounts of labels requires a heavy workload, and the uncertainty of the interpretation and labeling process decreases the model’s ability for making accurate predictions. Using the chalk-flint sequence scenario onshore Denmark as an example, we have developed a novel workflow for predicting subresolution thin layers from seismic sections. It entails generating large quantities of synthetic training data with high-quality labels using stochastic geologic modeling, training a convolutional neural network based on the synthetic data set, and applying it to real seismic data. This is, to our knowledge, the first example of using deep learning to predict subresolution thin layers from seismic data based on geostatistically generated training images. It is shown that a neural network trained on synthetic data can predict a realistic number of subresolution flint layers from the real seismic data that have been collected from the Stevns region in Denmark, which has value for the understanding of the overall geologic characteristics of succession and engineering applications such as construction site evaluation.