Research On Regularized Migration Imaging Methods Of Multisource Blended Data

Multisource seismic acquisition technology is an efficient new technology in seismic acquisition and it is regarded as a great revolution to the conventional acquisition. The direct imaging of multisource blended data is the main developing trend of this field. However, the multisource direct conventional migration is usually not satisfactory because of the crosstalk contamination. Aiming at this problem, this paper proposes two regularized migration methods of blended data direct imaging based on previous research. Both of these methods are effective and can deliver high quality images.Multisource seismic acquisition technology obtains blended seismic recording by continuously receive signals from multiple sources that locate in different position with simultaneous or random shorter delay. Compared with conventional single source acquisition technology, multisource seismic acquisition is not limited by source excitation time interval, and can largely improve acquisition efficiency and save exploration costs.In multisource seismic acquisition, the blended shot records overlap in both the spatial and temporal domains, which leads to complex blended wave fields and creates great difficulties to the following migration imaging. There are two imaging methods about the blended data at this stage: perform the migration processing directly on the blended data without any pre-separation, or recovery the blended data to individual shot data, which is called "deblending", then apply standard migration processing to these deblended data. The first method has a high processing efficiency; however, the multisource direct migration is usually not satisfactory because of the crosstalk contamination. The second method has low efficiency and the effect of source separation has serious influence on imaging quality. All things considered, the direct migration methods are the key trends in the field of blended data imaging and crosstalk suppression technology is the research priority.Seismic migration can be seen as a linear inversion problem like Ax? y. We can obtain imaging results with higher accuracy and better amplitude preservation in the sense of least square. Migration methods based on inversion are called least square migration(LSM). LSM is the major method for blended data direct imaging now. It has many migration operators and more flexible, which make it more adapt to the requirements of multisource technology and blended data processing. LSM often uses iterative optimization algorithm. In every iteration, we should conduct forward modelling(demigration) and migration at least once, which makes its calculation quantity N times compared to conventional forward modelling and migration(N is the number of iterations). As the noise level of blended data is higher, LSM always iterates many times to obtain better imaging results. As the iteration time increases, the calculation quantity of LSM will increase multiply, which induces large amounts of calculation costs. In LSM, if we use efficient migration algorithm and reduce the modelling time length in every iteration, the calculation costs can be reduced. In conventional migration algorithm, phase-shift migration is not the most accurate method, but it has high operating rate, small memory usage, can easily process any vertical velocity layer model and has no angle limit. As a result, in this paper, we combine phase-shift migration operator with LSM, and provide least square phase-shift migration. We aim at reducing the calculation costs of LSM through improving the calculation efficiency of iterative migration operator, and improving the imaging accuracy by compensating the disadvantages of conventional phase-shift migration with migration iteration. Numerical examples verify that least square phase-shift migration has better imaging accuracy and calculation efficiency.Like most geophysical inversion problems, LSM is a typical ill-posed problem. When using it, we often conduct regularization to ensure the stabilization of the solving process. From the LSM estimation point of view, researchers often add quadratic functional to the error function of LSM as penalty term, i.e., L2 norm regularized migration, also conventional LSM. As using quadratic penalty item as constraint condition, L2 norm regularized migration has smooth filter effect to the solutions, and can easily lead the model to excessive smooth, so as to suppress the discontinuous characteristics of the image. And the denoising ability of L2 norm regularized migration is ordinary, which cannot suppress the crosstalk effectively when it is oversize. To reduce the smooth filter effect of L2 norm and improve its denoising ability, we provide two kinds of new regularized migration methods for blended data, i.e., sparse constraint based L1 norm regularized migration and total variation based TV norm regularized migration. We study the corresponding fast solving algorithms, i.e., the fast iterative shrinkage-thresholding algorithm(FISTA) and the fast iterative shrinkage-thresholding algorithm/fast gradient projection(FGP)(FISTA/FGP). L1 norm regularized migration has good imaging effect to the high sparse subsurface model, and can accurately reconstruct sparse seismic image. TV norm regularized migration can process the discontinuity effectively. It can reserve the edge information of the image and also consider the smooth denoising and edge reservation. Multiple numerical modellings on theoretical model indicate that the two regularized migration methods provided in this paper have higher calculation efficiency, can effectively suppress crosstalk and improve direct imaging quality with blended data.