基于自适应字典学习的可控震源数据谐波噪声压制方法

可控震源高效采集地震资料中谐波噪声压制是一难点,虽然一些传统的基于稀疏优化的方法能够压制数据中的谐波噪声,但是由于使用固定字典,无法自适应地匹配有效信号的波形,存在有效信号损伤较大的问题。为此,提出了一种基于形态成分分析自适应学习字典的谐波噪声压制方法,用于分离原始相关后地震资料中谐波噪声干扰。首先对原始数据进行单道截取及分块处理组成样本集,然后利用K奇异值分解(K-SVD)学习得到超完备字典,进而应用字典原子的振幅谱比将字典分类为有效信号子字典与谐波噪声子字典,最后应用形态成分分析(MCA)理论将所得的子字典分别用于重建谐波噪声和有效信号,实现压制谐波噪声的目的。合成数据与实际数据的应用结果表明,基于自适应字典学习的可控震源数据谐波噪声压制方法在保护地震有效信号的同时能够有效压制谐波噪声。此外,对比近炮点数据和远炮点数据的谐波噪声压制结果可以看到,该方法对有效信号的损伤小于固定字典谐波噪声压制方法,具有良好的保真性与鲁棒性。

Suppressing harmonic noise in vibroseis data is challenging.Some traditional methods,based on sparse optimization,can suppress harmonic noise;but because of the use of a fixed dictionary,they are unable to adaptively match the waveform of the effective signal and can damage valid signals.In this study,a morphological component-based method using an adaptive-learning dictionary for harmonic noise suppression was proposed.First,harmonic noise was separated from the original,post-correlation,seismic data.Single-channel interception and block processing were then performed on the original data to form a sample set,and K-SVD learning was used to obtain an over-complete dictionary.Next,the atomic spectrum ratio of the dictionary atoms was used to classify the dictionary into effective signal subdictionaries and a harmonic noise subdictionary.Finally,the morphological component analysis theory was adopted to reconstruct the harmonic noise and effective signals using these two subdictionaries,respectively,to suppress harmonic noise.Tests on synthetic and field seismic data demonstrated that this method can effectively suppress harmonic noise while preserving seismic signals.Furthermore,it caused less damage to the effective signal than methods based on the fixed dictionary,thus verifying the fidelity and robustness of the method.

国家自然科学基金面上项目(41774135,41974131)及国家重点研发计划课题(2017YFB0202902)共同资助。