Deep Learning-Based Multistep Deblending with Adjacent Common Receiver Gathers Constraint
Shaochong Lin, Biao Wang · 2024
Summary Blended acquisition has emerged as a pivotal technique in improving acquisition efficiency and reducing costs. The obtained data contains overlapping information from various sources, i.e., blending noise contamination, which makes research on deblending algorithms essential. Deep learning-based deblending algorithms have recently gained popularity due to the superior accuracy and efficiency. However, the relationship among different common receiver gathers (CRGs) is often neglected. Considering the signals of adjacent CRGs are similar while the corresponding blending noise is different, we propose a deep learning-based multistep deblending with adjacent CRGs constraint. Three adjacent pseudo-deblended CRGs are used to form a three-channel input to amplify the signal coherence of the middle CRG, and the corresponding unblended CRG of the middle one is treated as the desired single-channel output. Additionally, a multistep strategy with blending noise simulation-subtraction is implemented to gradually attenuate the blending noise while preserving the signal. The effectiveness of our proposed method is well-demonstrated through synthetic and field data examples, showing marked improvement over the traditional unconstrained method.