Deep-unrolling architecture for image-domain least-squares migration

A general deep-unrolling architecture is proposed to solve the ill-posed inverse problem in the geophysical community. We apply the DU architecture to image-domain least-squares migration, which can provide improved reflectivity estimates. Moreover, the DU architecture can regularize the inversion process and can produce a qualitative measure of the uncertainty. The zip file "DU-IDLSM.rar" is the code of Python for DU-IDLSM and "data.rar" is bin data for DU-IDLSM.本文提出一种通用深度展开（deep-unrolling）架构，以解决地球物理学领域中的不适定逆问题。我们将该架构应用于成像域最小二乘偏移任务，可获得更为精准的反射率估计结果。此外，该深度展开架构能够对反演过程进行正则化处理，同时可生成不确定性的定性度量指标。压缩包"DU-IDLSM.rar"为DU-IDLSM的Python实现代码，"data.rar"为DU-IDLSM所使用的二进制（bin）数据。