Enhancing GPR Multisource Reverse Time Migration With a Feature Pyramid Attention Network

Abstract

The reverse time migration (RTM) algorithm is widely recognized in ground-penetrating radar (GPR) imaging for its high-resolution capabilities. However, the algorithm involves multiple forward modeling making it computationally intensive and less efficient. This article presents a workflow designed to enhance computational efficiency while maintaining the accuracy of RTM imaging. This purpose is achieved by implementing a source encoding strategy that integrates random polarity and time shifts to build a supergather as a new independent excitation source. This approach aims to suppress the crosstalk artifact among integrated excitation sources within the supergather during wave propagation, which could otherwise impact imaging accuracy. Subsequently, by integrating the feature pyramid attention network (FPANet) to further suppress residual multisource crosstalk artifact, thereby enhancing the overall imaging quality of RTM. Evaluations on synthetic GPR data demonstrate the algorithm’s capability to improve computational efficiency without sacrificing imaging accuracy, thereby confirming its effectiveness. Supported by both laboratory and field GPR data, the algorithm’s widespread applicability is proven. In summary, the proposed workflow is expected to enhance imaging efficiency significantly, achieving a $2\times $ – $5\times $ speedup ratio without compromising the quality of imaging progress.