Combination of shearlet and deep neural network: A more efficient and convenient denoising method
Yuxing Zhao, Yue Li, Ning Wu, Shengnan Wang, Qiankun Feng · 2022
Noise reduction is an essential step in seismic data processing. In this paper, we are studying the combination of conventional methods and emerging deep learning methods for seismic data denoising. Based on the residual learning strategy, we use a convolutional neural network to denoise the seismic data in the shearlet domain, avoiding the complicated and tedious process of threshold selection. Moreover, Shearlet transform has the characteristics of sparse representation, scale division, and direction division, which is conducive to the extraction of signal and noise features by the convolutional neural network and reduces the training difficulty. Through the combination of shearlet transform and convolutional neural network, the advantages of conventional methods and emerging deep learning methods can be complementary. The experimental results show that the proposed method can not only effectively reduce the noise in the seismic data, but also can effectively restore the seismic events with weak energy.