Double Dictionary Learning for Seismic Random and Erratic Noise Attenuation

In seismic data processing, denoising is one of the important steps to identify the Earth’s subsurface layer information. The erratic noise attenuation is always challenging due to the unknown distribution of high-amplitude peaks over seismic data. In literature, the double sparsity dictionary learning (DSDL) methods were used for erratic and random noise attenuation. Here, analytical and adaptive transformations are performed sequentially to attenuate erratic and random noises. However, the DSDL technique leads to a high computational cost due to K-SVD. Therefore, we propose a double dictionary learning (DDL) method to denoise both random and erratic noise by preserving the signal features from seismic data. The method uses two parallel adaptive dictionaries for simultaneous denoising, and both dictionaries are concatenated further to form a comprehensive dictionary. The regularized K-SVD was used to update the dictionary and sparse coefficients for signal preservation. The DDL method effectively reduced the computational costs. The DDL method was applied to different synthetic and field datasets for denoising. The numerical results show that the proposed method provides a higher signal-to-noise ratio (SNR), lower mean-squared error (mse), and less signal leakage than existing state-of-the-art denoising methods.