Ground penetrating radar inversion method based on multi-scale feature fusion

Abstract

The application provides a ground penetrating radar inversion method based on multi-scale feature fusion, and relates to the technical field of ground penetrating radar inversion, and comprises the following steps: S1, acquiring real-time scanning data of a current monitoring node ground penetrating radar; S2, constructing an encoder through a plurality of cascaded multi-scale hybrid convolution modules, wherein the encoder is used to acquire a multi-scale feature map of the real-time scanning data; S3, processing the multi-scale feature map to obtain a global feature map; S4, fusing the global feature map with features in the encoder through a cross-scale gating feature pyramid to generate a final fused feature map; and S5, outputting a physical property parameter distribution map of an underground medium of the real-time scanning data, so that the automation of ground penetrating radar inversion from data processing to result output is realized, the inversion precision and efficiency are greatly improved, meanwhile, complex manual preprocessing / postprocessing is not needed, and the real-time detection demand of an engineering site is adapted.

Images