An Efficient Point Spread Function Inversion Method for Image-Domain One-Way Wave-Equation Least-Squares Migration

Abstract

Image-domain least-squares migration (LSM) has demonstrated promising potential in enhancing the spatial resolution of migration images effectively and efficiently. However, existing image-domain approaches are mostly based on a local-stationary assumption, which estimates a local-stationary deblurring filter to process the corresponding subsection of the migration image. The deblurring precision is not fine enough. A point spread function (PSF) deconvolution method has been proposed to improve the resolution of migration images on a point-wise bias. Nevertheless, the computational and storage costs, particularly during the PSF process, remain significant. To achieve high-resolution imaging with reduced costs, we propose a PSF inversion method for image-domain one-way wave-equation (OWE) LSM. Leveraging a deep-learning optimizer, we achieve a rapid convergence for inverting the PSF in the spatial domain. In addition, we introduce a rescaled loss function for the stabilization and acceleration of the PSF inversion process. The rescaled loss function also makes it possible to obtain decent deblurring results during the early stages of iterations. Through some synthetic and field dataset experiments, it can be determined that our proposed PSF inversion method can produce high-resolution images with reduced migration artifacts and balanced amplitude. Additionally, our proposed method boasts noniterative characteristics, high parallelizability, freedom from regularization, and reduced storage and computational overhead, rendering it efficient and well-suited for practical applications.