Statistics-Guided Dictionary Learning for Automatic Coherent Noise Suppression
Statistics-Guided Dictionary Learning for Automatic Coherent Noise Suppression
复制标题
DOI：
10.1109/tgrs.2020.3039738
复制
发表时间：
2020-12
影响因子：
8.2
通讯作者：
Yatong Zhou;Jian Yang;Hang Wang;Guangtan Huang;Yangkang Zhang
中科院分区：
文献类型：
--
作者：
Yatong Zhou;Jian Yang;Hang Wang;Guangtan Huang;Yangkang Zhang
相干地震噪声由于其与有用信号的形态模式相似而难以衰减。为了衰减相干噪声，在最先进的方法中应该利用特殊的先验知识，这会导致显著的不便。在这里，我们开发了一个自动的方法来衰减相干噪声的自适应字典学习算法的基础上。自适应字典算法可以学习信号和相干噪声的特征，并在字典原子中留下明显的形态差异。字典原子中的这些差异可以转换为统计差异，这些差异可以被测量，然后用于区分信号原子和噪声原子。我们评估了表征字典原子的几种统计度量及其区分信号和噪声原子的特性。我们发现峰度度量最能代表信号原子和噪声原子之间的差异，然后我们设计了一个基于峰度的滤波器来滤除那些高峰度原子及其相应的系数向量，以抑制相干噪声。合成和真实的数据的例子证明了所提出的算法的性能。
Coherent seismic noise is usually difficult to attenuate due to the similar morphological patterns between noise and useful signals. To attenuate coherent noise, special preknowledge should be utilized in a state-of-the-art approach, which causes significant inconvenience. Here, we develop an automatic method to attenuate coherent noise based on the adaptive dictionary learning algorithm. The adaptive dictionary algorithm can learn the features of both signals and coherent noise and leave obvious morphological differences in the dictionary atoms. These differences in the dictionary atoms can be transformed into statistical differences, which can be measured and then used to distinguish between signal and noise atoms. We evaluate several statistical metrics in characterizing the dictionary atoms and their feasibilities in distinguishing between signal and noise atoms. We find that the kurtosis metric can best represent the differences between signal and noise atoms, and then we design a kurtosis-based filter to reject those high-kurtosis atoms and their corresponding coefficient vectors for suppressing the coherent noise. Synthetic and real data examples demonstrate the performance of the proposed algorithm.