基于正则化局部自适应核回归的单点高密度地震资料随机噪声压制方法

单点高密度地震资料的波场信息丰富，但同时存在信噪比较低、有效信号严重混杂在噪声中的问题。经典核回归法可以在地震数据中同相轴连续的区域获得近似最佳的滤波效果，但在同相轴突变区域容易造成边缘模糊。为了更有效地处理地震数据，研究了正则化局部自适应核回归（RLASKR）方法进行随机噪声压制。传统核回归法将空间距离作为唯一的回归函数影响因素，而正则化局部自适应核回归方法综合考虑了空间距离和灰度距离，核函数的形状随着不同区域数据样本的特征而变化，因此地震记录中的同相轴边缘信息能够有效地保存下来。模型记录和实际地震数据测试都显示出该方法的灵活性和有效性，验证了该方法在振幅保真和噪声压制方面比传统核回归法有着更好的效果。

Random noise attenuation method of single-point high-density seismic data based on regularized locally adaptive steering kernel regression

Wave field information is abundant in single-point high-density seismic data, but there are serious problems such as low signal-to-noise ratio and effective signals mixed in noise. The classical kernel regression method can obtain the approximate optimal filtering effect in the region with continuous event of seismic data, but it is easy to cause the edge blurring in the region with seismic event abrupt. In order to deal with seismic data more effectively, the regularized locally adaptive steering kernel regression (RLASKR)method for random noise attenuation is developed in this study. The classical kernel regression method takes the space distance as the only influence factor of regression function. The RLASKR method synthetically considers the space distance and gray distance. The shape of the kernel function changes with the characteristics of data samples in different regions, so the seismic event edge information can be effectively preserved in the seismic record. Model record and actual seismic data show the flexibility and effectiveness of this method. It is proved that the method has a better effect than the classical kernel regression in terms of amplitude preservation and noise attenuation.