Abstract: Traditional single-source seismic exploration has problems of low efficiency and insufficient anti-interference ability. Although multi-source technology improves the exploration efficiency, the data quality deteriorates due to the interference of aliasing noise. For this reason, this paper proposes two optimization methods to solve the source separation problem. Method 1: A dynamic weighted hybrid optimization algorithm (ALFT) is constructed by integrating the FISTA algorithm and the ALBM algorithm. This algorithm improves the convergence speed while ensuring accuracy. By combining the advantages of the filtering method and the inversion method, a process of “initial value pre-judgment-iterative correction” is formed. The experimental results show that, compared with the direct iteration method, this method can increase the signal-to-noise ratio by 10%～25% and reduce the iteration time by 33%. Method 2: A CSA-Unet deep learning network model is proposed. Based on the U-Net network architecture, this model introduces an attention local contrast (ALC) module to enhance the ability to capture the characteristics of effective signals, and combines a local entropy discrete point suppression mechanism to eliminate the interference of auxiliary sources. The validation results demonstrate that CSA-UNet achieves a significantly higher separation signal-to-noise ratio than ALFT_a and U-Net on both the simulated dataset (Sigsbee2B) and the real dataset, while also effectively preserving the structure of the formation reflection signals. The methods proposed in this paper provide an efficient and high-precision solution for multi-source seismic exploration and are of great significance in imaging practices under complex geological conditions.

Key words: multi-source seismic exploration, source aliasing noise, separation of primary and auxiliary sources, U-Net