中文摘要：

叠前逆时偏移（RTM）求解全波动方程进行波场延拓,可以精确描述地震波的传播,在陡倾界面和复杂构造成像方面具有显著优势.但在实际应用中,互相关成像条件会带来低波数高振幅的假象.本文分析了假象产生的原因,实现并比较了Laplace滤波、坡印廷矢量法和上下、左右行波分解这三种假象压制方法,针对不同的计算平台分析其优劣.数值试验表明,坡印廷矢量法当波场简单时较为有效;而上下、左右行波分解法和Laplace滤波可以应对较复杂的波场.就去假象结果而言,使用波场分解成像条件并施加Laplace滤波效果最佳;然而就计算效率而言,CPU平台应用波场分解成像条件只增加少量的计算量,但GPU平台的实现则需要对数据的读写付出较大的代价.因此,综合考虑,对于CPU平台上述两种方案的组合为最佳策略;而基于GPU平台的逆时偏移采用Laplace滤波去噪仍然是目前最经济的选择.Marmousi模型和SEG/HESS VTI模型试验验证了上述结论.

英文摘要：

Prestack reverse time migration （RTM） solves full wave-equation for wavefield extrapolation so that propagation of seismic wave can be described precisely. It shows great advantage in imaging dip reflectors and complex structures. However, traditional cross correlation imaging condition brings low-wavenumber and highamplitude artifacts in its application. In this paper, we analyze the cause of the artifacts, then implement and make close comparisons among three methods on CPU and GPU platform respectively： namely Laplace filtering, Poynting vector and wavefield decomposition. Numerical test shows that Poynting vector works well only when the wavefield is simple. But Laplace filtering and wavefield decomposition can handle complex wavefield. Wavefield decomposition imaging condition plus Laplace filtering achieves satisfactory results but brings additional computational cost, especially on GPU platform. So considering efficiency and quality of the image, that combination is currently the best choice for CPU platform. But simply applying Laplace filter remains the most economical way of removing artifacts for GPU platform. Finally, we conduct numerical examples of Marmousi and SEG/HESS model to validate the conclusion above.