基于稀疏表达的OBS去噪方法

摘要

Noise suppression is the basis for OBS data interpretation and subsequent inversion.Combining Curvelet transform and compression sensing,we propose a noise suppression method for OBS data using sparse representation.Comparing to the wavelet method,the Curvelet transform has advantage in identifying linear anomalies on a parabolic scale,which permits to reconstruct OBS data in a sparse representation domain.The sparse data is enhanced and reconstructed by the means of compression sensing,followed by transforming the coefficient to get iterative filtered by the cooling threshold of L1,then an optimal coefficient is resolved.Our study shows iterative filtering in the Curvelet domain with a cooling threshold can be utilized innoise suppression of OBS data.Comparison of wavelet and Curvelet transforms shows that the Curvelet method has a better S/N ratio in the circumstance of the same amount iterations of noise suppression.We use this new method to enhance the OBS data signal acquired from the Bohai Bay experiment and show clearer identification of the seismic phases and better S/N ratios,which facilitates picking up seismic phases from data and subsequent inversion of velocity models.%海底地震仪(OBS)采集数据的去噪处理是开展OBS震相分析及后续处理反演的基础.本文结合曲波(Curvelet)变换及压缩感知提出一种稀疏化表达的OBS去噪方法,并通过与小波变化对比等探讨去噪效果.曲波变换具有抛物尺度及识别线性异常的优点,可以稀疏地表示OBS数据,再结合压缩感知思想对稀疏表达OBS数据进行去噪处理和重构.通过对变换后的系数进行基于L1范数的冷却阈值迭代滤波,获得最优的变换系数,本文指出基于曲波变换的冷却阈值迭代法能够很好地对OBS数据去噪.对比小波和曲波两种变换在相同迭代次数下对理论模型数据进行去噪处理,表明曲波变换得到的结果信噪比更高.利用本文方法对渤海地区采集的OBS数据进行去噪处理获得了更加清晰连续的震相,噪声压制效果更明显,为震相拾取及后续速度模型反演奠定了良好的基础.