Irregularly seismic data interpolation based on deep learning with integrated channel-spatial attention mechanism

To address the challenges of irregular sampling and insufficient spatial sampling in field seismic data, this study proposed a deep learning-based interpolation method incorporating dual channel spatial attention mechanisms (CSAM). The proposed model establishes a collaborative framework of channel and spatial attention, enhancing feature representation by establishing connections between local reflection characteristics and global structural features. The performance of the method was evaluated through synthetic data experiments, including sparsity sensitivity tests, noise sensitivity tests, and field data validation, using metrics such as signal to noise ratio (SNR), mean absolute error (MAE), and structural similarity index (SSIM). Comparative analyses were conducted with Fourier projection onto convex sets (Fourierpocs), the classic U-net, and the efficient channel attention U-net (ECAUnet). Results demonstrate that the proposed method outperforms existing methods in reconstructing seismic reflection events and preserving amplitude fidelity, particularly in scenarios with extensive random data missing.