模型约束下的在线字典学习地震弱信号去噪方法

本文针对噪声成分和噪声结构的复杂性及弱信号的特征，发展了最新的在线字典学习去噪方法.在线字典学习去噪方法是以数据驱动的方式，反复进行学习构建字典方式，求得信号的稀疏性解以实现对信号的去噪，在此基础上，提出了数据驱动与模型驱动联合的模型约束下的在线字典学习去噪方法，先通过模型驱动方式获得一个较优质的学习样本以构建字典再进行去噪处理.通过和传统小波变换进行理论地震合成记录的效果对比，在高噪声比例的弱信号情况下远远优于传统的时频域去噪方法.实际数据去噪处理表明，模型约束下的在线字典学习去噪方法是一种有效的去噪方法，这种联合去噪方式能在高噪声背景下有效地提取出弱信号，具有广阔的推广应用前景.

Online dictionary learning seismic weak signal denoising method under model constraints

In this paper, the latest online dictionary learning denoising method is developed for the complexity of noise components and noise structures and the characteristics of weak signals. The online dictionary learning denoising is conducted by means of data-driven and iterative learning to obtain the sparse solution of the signal to realize the denoising of the signal. Based on this, an online dictionary learning denoising method under the combined constraints of data-driven and model-driven models is proposed. A better quality learning sample is obtained in a model driven process to build a dictionary and then to conduct denoising. Compared with the traditional wavelet transform for theoretical seismic synthesis recording, it is far superior to the traditional time-frequency domain denoising method in the case of low-SNR weak signals. The actual data denoising process shows that the online dictionary learning denoising method under model constraints is an effective denoising method. This joint denoising method can effectively extract weak signals against high noise and has broad application prospect.