Random noise attenuation in seismic data using an adaptive thresholding and the second-order variant time-reassigned synchrosqueezing transform

Seismic data analysis often faces the challenge of random noise contamination from various sources. To overcome this, innovative noise attenuation methods utilizing seismic signal properties are needed. This study focuses on efficiently suppressing random noise in the domain of time and frequency by accurately estimating instantaneous frequency using the single-valued group delay characteristic of seismic signals. The time-reassigned synchrosqueezing transform (TSST) and its second-order variant (TSST2) offer high-resolution time-frequency representations (TFRs) for noise suppression. Expanding on these advancements, we propose an efficient noise suppression method that integrates the adaptive thresholding model into the TSST2 framework and employs sparse representation of the TFR through low-rank estimation. This method effectively attenuates noise while preserving essential signal information. The proposed approach operates trace by trace on recorded data, initially transforming it into a sparse subspace using TSST2. The adaptive thresholding model then decomposes the resulting TFR into sparse and semi-low-rank components, achieving a high-resolution and sparse TFR for efficient separation of noise and signal. After noise suppression, the seismic data can be fully reconstructed by inversely transforming the semi-low-rank component data into the time domain. This method addresses previous limitations in noise attenuation techniques and provides a practical solution for enhancing seismic data quality.