利用改进的在线字典学习估计时变子波  

Time-varying wavelet estimation based on improved online dictionary learning

作　　者：孔德辉[1,2] 彭真明[1,2] 

机构地区：[1]电子科技大学光电信息学院,四川成都610054 [2]电子科技大学信息地学研究中心,四川成都610054

出　　处：《石油地球物理勘探》2016年第5期901-908,835,共8页Oil Geophysical Prospecting

基　　金：国家自然科学基金(41274127;40874066)资助

摘　　要：为了获得符合实际的混合相位子波,提出了一种基于在线字典学习的时变子波估计方法。将时变子波估计转化为在线字典学习问题,通过过完备字典的在线学习实现冗余字典的自适应更新。字典中的每个原子代表子波的一个分量,通过原子的线性组合实现对时变子波的有效逼近。在线字典学习可以灵活地利用训练数据,改进字典中的原子,提升字典的自适应特性。同时,根据地震数据的特点,对训练数据与稀疏表示的残差项进行滤波处理,改进了在线字典学习方法,降低了对噪声的敏感性。无噪声和含噪声合成数据的实验结果证明了本文方法的有效性,而且对噪声具有一定的鲁棒性。实际子波估计结果以及Wiener滤波的反褶积剖面和频谱分析表明,本文方法得到的结果并未使噪声能量增强,但频带得到拓宽,从而为时变子波估计提供了新思路。In order to obtain a mixed phase wavelet which is fit with real state, we introduce a time-varying wavelet estimation method based on online dictionary learning. The time-varying wavelet estimation is transformed into an online dictionary learning problem where the redundant dictionary is adaptively updated through its online learning process. Each atom in the dictionary represents a component of the varying wavelet. Its accurate approximatation is realized by a line combination of those atoms. Online dictionary learning uses flexibly the training data, updates the atoms in the dictionary, and enhances the adaptability of the dictionary. According to characteristics of seismic data, residuals between the training data and sparse representation are filtered which improves the online dictionary learning and reduces the sensitivity to noise. In noise free and noisy synthetic data tests, the results show the validity of the proposed method and its robustness against noise. Wavelet estimation on field data, Wiener filtering deconvolution sections and spectral analysis are also obtained. The outcome shows that the proposed method widen frequency band without noise energy increase, which provides an alternative way to estimate time-varying wavelet. © 2016, Editorial Department OIL GEOPHYSICAL PROSPECTING. All right reserved.

关 键 词：在线字典学习 稀疏表示 训练集 滤波 反褶积 时变子波估计 

分 类 号：P631[天文地球—地质矿产勘探]