Swell noise removal in exploration geophysics is an essential and the first step in the processing sequence of marine seismic data processing. However, removing this contamination is challenging due to the characteristics varying along with the survey. Various machine-learning-based methods have been proposed to solve this problem. However, most of the available machine-learning methods employ the supervised learning method, where a good training dataset with many samples is required. Therefore, we propose a self-supervised learning method to reconstruct contaminated traces’ signals based on the information of adjacent clean traces and adjacent clean spectrum. The proposed method combines a blind-trace network, an automatic spectrum suppression technique and a trust spectrum boosting technique for high-accuracy seismic data reconstruction to denoising processing. Experiments on synthetic swell noise and the Sigsbee dataset demonstrate the proposed method’s effectiveness.
Note: This paper was accepted into the Technical Program but was not presented at IMAGE 2022 in Houston, Texas.
