SELF-GUIDED ATTENTION DENOISING NETWORK FOR PRE-STACK SEISMIC DATA: FROM COARSE TO FINE

摘要

Background noise attenuation is one of the most essential steps in seismic dataprocessing. Residual background noise is likely to cause some artifacts in the followingseismic imaging, thus bringing huge difficulties to the final interpretation. In recent years,deep-learning (DL) methods based on data driven strategy, especially the convolutionalneural network (CNN), work well in seismic noise attenuation. In addition, it is appliedautomatically without parameter fine-tuning after training. To further improve theirperformance, we propose a novel architecture: self-guided attention network (SGA-Net)by combining self-guided strategy and spatial attention mechanism. Different from mostof the conventional CNNs, this proposed SGA-Net can capture multi-scale features byperforming the convolution operation on seismic data with different resolutions. In thisnetwork, the self-guided strategy is adopted to take full advantage of the multi-scalefeatures; specifically, we utilize the global coarse features extracted at low resolution toguide the extraction process of local finer features at higher resolution. Furthermore, wedesign a spatial attention module with two inputs to fuse the global coarse and local finefeatures. We set up four competitive methods for SGA-Net including two traditionalseismic denoising methods and two existing DL denoising methods in both synthetic andreal experiments and experimental results demonstrate the advantage of SGA-Net both innoise attenuation and signal preservation.