Seismic Random Noise Suppression Based on Block Flattening BM3D

ABSTRACT

During the seismic data acquisition, the signal and random noise are superimposed on each other, which has a negative impact on the accuracy of subsequent imaging and inversion. Block matching and 3D collaborative filtering (BM3D) use similar redundant structures in the image to achieve noise suppression. However, for data with complex geological structures, their self‐similarity is disrupted, resulting in a reduction in redundant information. When processing such data, BM3D is prone to the problem of signal energy leakage. To solve this problem, we improve the BM3D algorithm by introducing a slope‐guided mechanism and propose a block flattening BM3D (BFBM3D) method. This method introduces a block flattening operator into the BM3D framework to enhance local self‐similarity. We adjust the events according to the slope information and flatten them horizontally, thereby reducing the inconsistency caused by varying slopes. The flattened data blocks are processed by three‐dimensional collaborative filtering, and finally the original shapes of the events are restored via inverse flattening to obtain the denoised result. The test results on synthetic and field data show that the proposed method can not only suppress noise efficiently but also preserve the signal energy well.