Automated seismic semantic segmentation using Attention U-Net

Seismic facies mapping from a three-dimensional seismic cube is of significant value to various seismic interpretation and characterisation tasks. Traditional facies mapping is based on examining sedimentary environments and stratigraphic sequences, that provide distinct characteristics used for facies mapping. Given the complex nature of the task, manual facies mapping is typically time and labour consuming, and the quality of the decisions vary as a function of expertise. This complexity is further increased with the ever-increasing size of 3D seismic datasets. Deep learning methods have shown a promising potential to perform fast, accurate and automated segmentation tasks. We investigate the application of machine learning techniques, particularly state-of-the-art deep convolutional neural networks (CNNs), as a framework to perform accurate automated seismic facies pixel-wise segmentation. The workflow consists of a CNN based U-Net architecture that adopts modern computer vision techniques. We propose three major changes to the standard U-Net to boost the performance for seismic semantic segmentation tasks: (1) using residual building blocks in the encoder (2) using transformer like attention gates after each residual block, and (3) using frequency spectrum data, in addition to seismic amplitude, as input to the network. We show that this implementation achieves higher accuracy metrics outperforming recently published state-of-the-art benchmarks. The performance of the proposed method is validated using two 3D seismic datasets, the F3 Netherlands dataset and the Penobscot dataset acquired offshore Nova Scotia, Canada. Experimentation involves training on a set of samples and tuning the hyper-parameters, followed by quantitative evaluation of the trained network. The proposed workflow produced high quality segmentation with significantly reduced artifacts, improved edge detection, and improved lateral consistency throughout the seismic survey.