Seismic Facies Analysis: A Deep Domain Adaptation Approach

Deep neural networks (DNNs) can learn accurately from large quantities of labeled input data but often fail to do so when are scarce. DNNs sometimes generalize on test sampled different distributions. Unsupervised deep domain adaptation (DDA) techniques have been proven useful no labels available and distribution shifts observed in the target (TD). In this study, experiments performed seismic images F3 block 3-D dataset offshore Netherlands [source (SD)] Penobscot survey Canada Three geological classes SD TD that similar reflection patterns considered. A DNN architecture named EarthAdaptNet (EAN) is proposed semantically segment few scarcity, we use a transposed residual unit replace traditional dilated convolution decoder block. The EAN achieved pixel-level accuracy >84% an $\sim 70$ % for minority classes, showing improved performance compared existing architectures. addition, introduce correlation alignment (CORAL) method create unsupervised network (EAN-DDA) classification reflections demonstrate possible approaches unavailable. Maximum class was 99$ 2 with overall >50%. Taken together, EAN-DDA has potential classify facies high accuracy.