Abstract: Existing noise reduction methods for seismic signals can not achieve a good result for the curve events of seismic profile. This paper proposes a noise suppression method for seismic signals based on Locally Linear Embedding（LLE） reconstruction and SVD decomposition. First, the method uses the reconstruction of locally linear embedding to reconstruct each sample of seismic data by its neighborhoods, which achieves the linearization of the curve events of nonlinear data. This removes the random noise in certain extent. Then, it separates the effective signals from the data containing the noise according to a good coherence of effective frequency signals for seismic data of LLE reconstruction, and gets rid of non-relevant noise. It ultimately achieves random noise suppression of seismic data. Finally, the experiments are conducted on forward model and real seismic data. The proposed method effectively removes random noise compared with traditional SVD method, and it does not lose the effective signals mainly.

Key words: Locally Linear Embedding（LLE）, Singular Value Decomposition（SVD）, reconstruction, decomposition, seismic signals, noise reduction

摘要： 针对现有地震信号降噪方法处理地震剖面的弯曲同相轴效果不佳，提出联合局部线性嵌入（LLE）和奇异值分解（SVD）方法的地震信号降噪技术。利用LLE的重构思想，对地震数据采样点用其近邻进行重构，实现非线性模式的弯曲同相轴的线性化处理，并去除一定程度的随机噪声；根据地震资料有效信号具有良好相关性的特性，采用SVD分解对LLE重构后的地震数据进行有效信号和噪声分离，剔除不相干的噪声，最终实现地震数据的随机噪声压制。在正演模型和真实地震资料上进行了实验，实验结果表明：与传统SVD方法相比，提出的方法很好地消除了随机噪声，有效信号基本上无丢失。

关键词: 局部线性嵌入, 奇异值分解, 重构, 分解, 地震信号, 去噪