C2S-net: A confidence-guided multi-stage network for seismic data consecutively missing trace interpolation

Consecutively missing traces pose a major challenge for seismic data reconstruction. Although multistage neural networks are a promising approach, each stage often inherits reconstruction errors from the previous one. As a result, seismic events that are poorly reconstructed in early stages remain largely unrecoverable in subsequent stages, limiting the performance of conventional multi-stage strategies under severe consecutive missing conditions. To address this issue, we propose C2S-Net, a confidence-guided multi-stage network specifically designed for seismic data with consecutively missing traces. C2S-Net introduces two key mechanisms that enhance the multistage reconstruction process. First, a difficult sample generation module (DGM) is employed to explicitly constructs samples that are particularly challenging for the first stage to recover. This forces later stages to focus on the areas that are most prone to failure. Second, a confidence module (CM) estimates pixel-level reconstruction reliability, progressively narrowing the missing region and allowing subsequent stages to prioritize weak or ambiguous events. Experimental results on both synthetic and field data demonstrate that C2S-Net outperforms single-stage and conventional multistage methods in terms of structural fidelity, weak-reflection preservation, and overall reconstruction quality.