基于稀疏反演的同步震源地震数据分离方法

同步震源混合地震数据的有效分离方法是同步震源采集技术得到广泛应用的关键之一。为此,在时间随机抖动的混合采集框架下,提出了一种同步震源混合地震记录分离方法。首先将未混合的共炮点记录作为训练样本,基于字典学习,得到学习型字典;然后基于稀疏反演构造分离混合地震记录的反问题表达式,并将待分离记录基于学习型字典的稀疏表示作为约束项,利用交替迭代策略进行求解。复杂模型数据和实际资料算例表明,与基于局部离散余弦变换的稀疏反演结果相比,利用字典学习能够有效提高分离结果的精度,为后续的地震信号处理提供高质量的数据。

Separation of blended seismic data acquired in simultaneous source acquisition is substantially necessary.A sparse inversion-based method to separate blended data in case of a random time-dithering scheme is presented in this paper.The first step is the extraction of block training data from clean shot gathers without blending.The extracted data is used to train a learned dictionary through the K-SVD algorithm,based on sparse representation and patch-wise dictionary learning.An inverse problem expression for the separation of blended data was then developed based on sparse inversion.We used the sparse representation of blended data as regularization constraint and performed an alternate iterative scheme to update the separated recovery data and sparse coefficients respectively.Testing on the synthetic and field data demonstrated that the recovery data obtained from dictionary learning had better separation accuracy compared to that based on two dimensional fixed local discrete cosine transform.

国家自然科学基金项目(41504093,41774135)和陕西省工业攻关项目(2015GY058)共同资助。