基于点扩散函数的图像学高分辨率技术

常规基于一维假设的高分辨率处理技术如反褶积、反Q滤波方法等, 针对水平构造或者小倾角构造有较好的处理效果, 但在面向高陡构造或者断裂等复杂构造时处理效果不佳。而偏移成像结果可以看作是点扩散函数(PSF)与真实反射系数褶积的结果, 高维空间条件下的高分辨率处理技术实际上是此正演问题的一个反问题, 即基于成像结果进行反射系数反演。为此, 提出了一种基于点扩散函数的高分辨率处理技术, 首先基于点扩散函数的计算原理, 推导出了基于高频近似条件下的快速求解算法, 并在高维空间褶积理论的指导下结合图像学反演算法研发了纯数据驱动的点扩散函数提取技术, 最后采用高维空间反褶积算法求得高分辨率成像结果, 实现地震频带的有效拓宽和成像分辨率的提升。模型数据和实际数据测试结果表明, 该算法可在不损失低频的前提下有效扩宽高频, 且具有在计算效率与常规高分辨率算法基本相当的条件下有效提高不同展布方位地质体分辨率的优势。

The focus of seismic exploration has gradually shifted from structural reservoirs to concealed reservoirs, which raises high requirements for high-resolution processing technology.Conventional high-resolution processing techniques, such as deconvolution and inverse Q filtering, are usually based on one-dimensional assumptions and are effective for horizontal structures or small dip-angle structures.However, these techniques are invalid theoretically for complex structures such as high-steep structures or faults.As we know, the migration image is the convolution of point spread function (PSF) and true reflection coefficient.The high-resolution processing technology in high-dimensional space is actually an inverse problem of this forward modeling problem, which is to solve the reflectivity based on migration result.In this paper, the high-resolution processing technology of imagery based on PSF is proposed.Firstly, a fast algorithm for solving the PSF in high-frequency approximation is deduced according to the computing theory of PSF.Then, a purely data-driven PSF extraction technique is developed depending on the theory of deconvolution in high-dimensional space together with the imagery.Finally, the high-dimensional space deconvolution algorithm is used to obtain high-resolution imaging result, so that the frequency band is widened.The tests of model and field data show that the algorithm proposed in this paper can effectively expand high frequencies without losing low frequencies and has the advantage of improving the resolution of geological bodies with different distribution azimuths compared with the conventional high-resolution algorithms.