A robust random noise suppression method for seismic data using sparse low-rank estimation in the time-frequency domain

The noise separation from seismic data is of significant importance in geophysics. In most cases, the random noise always overlaps the seismic reflections over time, which makes it challenging to suppress. To enhance seismic signal, we propose a robust noise suppression method based on high-order synchrosqueezing transform (FSSTH) and robust principal component analysis (RPCA). Firstly, the noisy seismic data is transformed into a sparse time-frequency matrix (TFM) using the FSSTH. Then, the RPCA algorithm is employed to decompose the sparse TFM into a low-rank matrix and a sparse matrix that can be used to depict the useful signal and noise, respectively. Finally, the denoised signal can be obtained by back-transforming the low-rank matrix to the time domain via the inverse FSSTH. We utilize a synthetic data and two field datasets to demonstrate the robustness and superiority of our method, and compare with the conventional denoising algorithms such as $f-x$ denconvolution and $f-x$ singular spectrum analysis (SSA). The obtained results indicate that the proposed method is capable of achieving an excellent tradeoff between random noise suppression and seismic signal preservation.