基于字典学习的地震数据随机噪声压制算法

Random Noise Suppression Algorithm for Seismic Data Based on Dictionary Learning

针对传统变换基函数难以获得地震数据最优的稀疏表示,提出基于字典学习的随机噪声压制算法,将地震数据分块,每一块包含多个地震记录道在一定采样时间段内波形的信息,利用自适应字典学习技术,以地震数据块为训练样本,根据地震数据邻近块中记录道相似的特点,构造超完备字典,稀疏编码地震数据,从而恢复数据的主要特征,压制随机噪声.实验表明算法具有较高的PSNR值,并且能较好的保持地震数据纹理复杂区域的局部特征.

Aiming at the problems that the traditional sparse transform functions are difficult to obtain the optimal representation of seismic data, thus the random noise suppression algorithm based on dictionary learning is proposed. The seismic data are divided into several blocks, and each block contains multiple seismic records in a certain period time of sampling waveform ：information, the over-complete dictionary learning technique is used with the seismic data blocks as the training samples. According to the characteristics of the seismic data, over-complete dictionary is adaptively constructed, and the seismic data are sparsely coded, so as to restore the main features of the data, and then to suppress random noise. The experimental results show that the algorithm yields higher PSNR, and preserves the local characteristics of the complex region of seismic data better.