Seismic data interpolation by greedy local Radon transform
Seismic data interpolation by greedy local Radon transform
复制标题
DOI：
10.1190/1.3484195
复制
发表时间：
2010-11
期刊：
影响因子：
3.3
通讯作者：
Juefu Wang;M. Ng;M. Perz
中科院分区：
文献类型：
--
作者：
Juefu Wang;M. Ng;M. Perz
提出了一种空间局域高分辨率Radon变换的贪婪逆方法。该方法的核心是基于传统的迭代算法--共轭梯度(CG)，但在幅度优先的局部模型空间中自适应地使用。自适应反演引入了一种面向相干的机制来增强重要模型参数的聚焦，从而提高了模型的分辨率和收敛速度。我们采用时空域局部线性Radon变换的思想进行数据内插。我们发现局部Radon变换涉及迭代地应用空间局部化的向前和伴随Radon算子来拟合输入数据。通过促进模型稀疏性的子空间算法可以找到最优的局部Radon面板，并且可以使用得到的局部Radon面板来预测丢失的数据。子间隔策略大大降低了计算局部Radon系数的代价，从而降低了总的反演代价。该方法可以处理不规则和规则的几何图形以及明显的空间走样。我们用三个简单的合成数据集与最小加权范数傅立叶插值法比较了该方法的性能，并展示了新算法在空间混叠数据内插的优势。我们还在2D合成数据和野外数据集上对算法进行了测试。这两个测试都表明，该算法是一种稳健的抗锯齿工具，尽管它不能完全恢复丢失的强弯曲事件。
We propose a greedy inversion method for a spatially localized, high-resolution Radon transform. The kernel of the method is based on a conventional iterative algorithm, conjugate gradient (CG), but is utilized adaptively in amplitude-prioritized local model spaces. The adaptive inversion introduces a coherence-oriented mechanism to enhance focusing of significant model parameters, and hence increases the model resolution and convergence rate. We adopt the idea in a time-space domain local linear Radon transform for data interpolation. We find that the local Radon transform involves iteratively applying spatially localized forward and adjoint Radon operators to fit the input data. Optimal local Radon panels can be found via a subspace algorithm which promotes sparsity in the model, and the missing data can be predicted using the resulting local Radon panels. The subspacing strategy greatly reduces the cost of computing local Radon coefficients, thereby reducing the total cost for inversion. The method can handle irregular and regular geometries and significant spatial aliasing. We compare the performance of our method using three simple synthetic data sets with a popular interpolation method known as minimum weighted norm Fourier interpolation, and show the advantage of the new algorithm in interpolating spatially aliased data. We also test the algorithm on the 2D synthetic data and a field data set. Both tests show that the algorithm is a robust antialiasing tool, although it cannot completely recover missing strongly curved events.