Seismic data restoration is an ill-posed inverse problem. Based on the sparseness of seismic data in the curvelet domain, this problem can be transformed into a sparse optimization problem. This paper proposes to use the approximation of zero-norm as the objective function and develop a projected gradient method to solve the corresponding minimization problem. We also employ a recently proposed piecewise random sampling method which can control the sampling gap and keep the randomness of sampling. Numerical results show that the projected gradient method can reduce the amount of computation greatly, and the restoration based on the piecewise random sampling is better than that of random sub-sampling.
One of reverse time migration’s main limitations is that an unscaled adjoint operator is prone to produce images with low resolution, inaccurate amplitudes, and even artifacts. Least-squares reverse time migration (LSRTM) has been introduced to mitigate this inadequacy via the use of an approximation to an inverse operator. LSRTM suffers from its own limitations, most importantly from poor condition, which often manifests itself as image artifacts. One approach to ameliorate this issue is to constrain the optimization problem by introducing a penalty term to the cost function. Penalizing estimated parameters for sparsity is one such constraint that has been shown to be effective. A drawback of this technique is that it introduces a trade-off between data fitting and image sparsity. Furthermore, if using the Cauchy constraint, an additional trade-off is introduced due to the requirement to estimate a hyperparameter. We introduce an alternative approach that mitigates these trade-offs by combining a multiplicative cost function with an effective means for determining the Cauchy hyperparameter. We also introduce a new formulation of the multiplicative cost function that avoids over-penalization by the constraint via the introduction of a relaxation term. Finally, we seek to improve the computational efficiency by introducing a new approach for computing the step length. As such, our method introduces three novel aspects to constrained LSRTM: (1) a relaxed multiplicative cost function, (2) semiautomatic estimation of the Cauchy hyperparameter, and (3) efficient computation of the step length. We discuss the theory and implementation, followed by application to three synthetic data sets and a real ultrasonic data set. Given the presence of large salt bodies, elasticity, and noise, along with the directivity of piezoelectric ultrasonic transducers, these data sets provide a challenging test of the approach outlined. Results demonstrate that our method is robust in handling the challenges imposed by these scenarios.
Exploration Geophysics is published by CSIRO PUBLISHING on behalf of the Australian Society of Exploration Geophysicists. Papers published report the results of significant case histories and relevant original research in geophysics, with emphasis on the Australian and similar environments.
Least-squares reverse time migration (LSRTM) is currently one of the most advanced migration imaging techniques in the field of geophysics. It uses least-squares inversion to fit the observed data, resulting in high-resolution imaging results with more accurate amplitudes and better illumination compensation than conventional reverse time migration (RTM). However, noise in the observed data and the Born approximation forward operator can result in high-wavenumber artifacts in the final imaging results. Moreover, iteratively solving LSRTM leads to one or two orders of computational cost higher than conventional RTM, making it challenging to apply extensively in industrial applications. Simultaneous source acquisition technology can reduce the computational cost of LSRTM by reducing the number of wavefield simulations. However, this technique also can cause high-wavenumber crosstalk artifacts in the migration results. To effectively remove the high-wavenumber artifacts caused by these issues, simultaneous source and deep learning are combined to speed up LSRTM as well as suppress high-wavenumber artifacts. A deep residual neural network (DR-Unet) is trained with synthetic samples, which are generated by adding field noise to synthesized noise-free migration images. Then, the trained DR-Unet is applied on the gradient of LSRTM to remove high-wavenumber artifacts in each iteration. Compared to directly applying DR-Unet denoising to LSRTM results, embedding DR-Unet denoising into the inversion process can better preserve weak reflectors and improve denoising effects. Finally, the proposed LSRTM method is tested on two synthetic data sets and a land data set. The tests demonstrate that the proposed method can effectively remove high-wavenumber artifacts, improve imaging results, and accelerate convergence speed.
Exploration Geophysics is published by CSIRO PUBLISHING on behalf of the Australian Society of Exploration Geophysicists. Papers published report the results of significant case histories and relevant original research in geophysics, with emphasis on the Australian and similar environments.