应用GPU的傅里叶有限差分逆时偏移

逆时偏移是针对复杂构造的主流叠前深度偏移成像技术，常规逆时偏移技术大多采用高阶有限差分和成像后滤波算法。由于实际偏移的计算网格较大造成空间频散现象和成像后的去低频噪声滤波会损害有效低频信息，因此成像频宽受到一定影响。为此，提出基于GPU的傅里叶有限差分逆时偏移方法。即采用傅里叶有限差分算法进行波场传播计算，波场外推中直接在时间—波数域做显式波场分解和互相关成像；同时采用基于GPU的算法，显著提高了傅里叶有限差分逆时偏移的计算速度。实际资料应用结果表明，该技术能实现对高、低频有效信息的充分保护和利用，从而提高了地震成像分辨率。

Fourier finite-difference reverse time migration using GPU

Reverse time migration (RTM) is a pre-stack depth migration technique commonly used for imaging complex structures.In conventional RTM,high-order finite difference and post-imaging filter-ing algorithms are usually used，but spatial dispersion will be caused by large computing girds and effective low-frequency information will be lost when using post-imaging filtering to remove low-frequency noises.A GPU-based Fourier finite-difference RTM method is developed.It uses the Fourier finite-difference algorithm to calculate wavefield propagation,and explicit wavefield decomposition and cross-correlation imaging are directly performed in time-wavenumber domain du-ring wavefield extrapolation.In addition,the GPU-based algorithm can significantly improve the calculation speed of Fourier finite difference RTM.The application in field data shows that the method can protect and utilize the high and low frequency information effectively，so the imaging resolution can be improved apparently.