基于U-Net网络的二维小波域地震数据随机噪声衰减

基于深度学习的地震数据噪声衰减方法比传统去噪方法更加高效, 去噪结果具有更高的信噪比。现有基于深度学习的去噪方法通常在时空域对地震数据进行处理, 但小波域中有效信号与噪声之间的特征差异更为明显, 有利于网络训练学习及噪声衰减。利用二维小波域地震数据的稀疏性和多尺度性, 联合二维离散小波变换与U-Net网络, 提出了基于U-Net网络的二维小波域随机噪声衰减方法(Dwt-U-Net)。该方法先对地震数据进行二维离散小波变换, 再以二维小波系数作为网络输入和输出进行网络训练, 获得去噪后的小波系数, 最后将该小波系数进行重构得到去噪结果。模拟数据和实际地震数据测试及与不同方法对比结果显示, 在不同噪声水平情况下, Dwt-U-Net方法的去噪结果具有更高的信噪比和保真度。此外, 相对于传统时空域U-Net网络去噪方法, Dwt-U-Net方法在提高信噪比的同时, 网络训练时间减少一半左右。

Compared with traditional denoising methods, the deep-learning-based noise attenuation method for seismic data is more efficient, and the denoising result has a higher signal-to-noise ratio.Existing denoising methods based on deep learning typically process seismic data in the time and space domains.However, the feature difference between the effective signal and noise in the 2D wavelet domain benefits network training and noise attenuation.Taking advantage of the sparsity and multiscale nature of seismic data in the 2D wavelet domain combined with the 2D discrete wavelet transform and U-Net network, this study proposed a random noise attenuation method in the 2D wavelet domain based on the U-Net network (Dwt-U-Net).This method first performed a 2D discrete wavelet transform on seismic data, then used 2D wavelet coefficients as the network input and output to train the network to obtain denoised wavelet coefficients, and finally reconstructed the wavelet coefficients to obtain denoising results.In this study, simulations and actual seismic data were tested to verify the accuracy of this method.Compared to different traditional methods, the test results have shown that the denoising results of the Dwt-U-Net method had a higher signal-to-noise ratio and fidelity under different noise levels.Compared with the traditional U-Net network denoising method, the proposed method reduced the network training time by approximately half while improving the signal-to-noise ratio.