基于高精度字典学习算法的地震随机噪声压制

地震勘探中的随机噪声会对有效信号产生严重干扰,甚至会导致信号畸变.为了满足高精度的地震勘探要求,应用高精度字典学习算法压制随机噪声.该算法基于超完备稀疏表示理论,对传统的字典学习算法从字典训练和稀疏编码两个方面进行了改善:在字典训练阶段,保持支集完备的情况下,对字典进行循环更新,使字典更好地适应数据;在稀疏编码阶段,选用上一轮更新的部分大系数作为新一轮迭代的初始系数,充分利用大系数表示有效信号的特点.对理论数据和实际含噪地震资料的处理结果表明,与传统算法相比,高精度字典学习算法在去噪的同时能保护有效信息,明显改善去噪精度,对地震记录信噪比的提高有显著优势.

Seismic random noise suppression based on the high-precision dictionary learning algorithm

In seismic exploration,the random noise severely distorts and interferes with seismic signals,and hence the denoising process is very important.In order to meet the high-precision requirement,the authors,based on the sparse and redundant representation theory,improve the dictionary update stage and the sparse coding stage in the conventional dictionary learning algorithm.While keeping the supports intact,the dictionary atoms are recurrently updated to adapt them to the specific seismic data.In the dictionary domain,large coefficients represent effective signals.Taking full advantage of this characteristic,the authors use several large coefficients from the last round of iteration as initial coefficients.In this way,the computational efficiency of the learning algorithm can be improved.The new algorithm is applied to synthetic and field seismic records and compared with the conventional K-SVD algorithm.The denoising results are satisfactory.It is shown that the new method can remove the random noise and protect the effective information at the same time.It is competitive in improving the signal-to-noise ratio of seismic records.