Unsupervised Deep Learning Framework for 5D Seismic Denoising and Interpolation

We propose an unsupervised framework to reconstruct the missing data from the noisy and incomplete five-dimensional (5D) seismic data. The proposed method comprises two main components: a deep learning network and a projection onto convex sets (POCS) method. The model works iteratively, passing the data between the two components and splitting the data into a group of patches using a patching scheme. Specifically, the patching scheme breaks the input data into small segments which are then reshaped to a vector of one dimension feeding the deep learning model. Afterward, POCS is utilized to optimize the output data from the deep learning model, which is proposed to denoise and interpolate the extracted patches. The proposed deep learning model consists of several blocks, that are, fully connected layers, attention block, and several skip connections. Following this, the output of the POCS algorithm is considered as the input of the deep learning model for the following iteration. The proposed model iteratively works in an unsupervised scheme where labeled data is not required. A performance comparison with benchmark methods using several synthetic and field examples shows that the proposed method outperforms the traditional methods.