Seismic resolution enhancement with self-supervised learning Available to Purchase

The concept of neural network (NN)-based seismic resolution enhancement has gained a lot of traction recently. Yet, the majority of work rely on training NNs on synthetic data via a supervised learning strategy, often encountering generalization issues on real data. To address this problem, we develop a self-supervised learning method for seismic resolution enhancement. Specifically, we reinterpret seismic resolution enhancement as a frequency extension task, particularly focusing on the reconstruction of high-frequency components. Initially, we warm up the NN using the original bandwidth-limited data as pseudo labels, with input data derived by filtering out highfrequency elements from the original data. Subsequently, the network undergoes iterative data refinement, where pseudo labels are predicted from the NN trained in the previous epoch on the original data, and input data are obtained by filtering high-frequency components from these predictions. The efficacy of our method is demonstrated through tests on both synthetic and field data.