Recovering 3D Salt Dome by Using Gravity Data Inversion Using ResU-Net++

In myriad disciplines such as mineral exploration, geological survey, groundwater resource inspection, and environmental surveillance, gravity inversion is a method ubiquitously employed. However, conventional gravity inversion approaches grapple with formidable challenges, including heightened susceptibility to minute data variations and the propensity for descent into numerous local minima of the error function. To mitigate these conundrums, we explore deep learning methodologies, specifically presenting the ResU-Net++, a network that synergistically integrates residual connectivity and deep feature fusion tactics. We scrutinize the efficacy of this novel model via an array of simulation experiments encompassing four distinct networks: the AttU-Net, the R2U-Net, the Nested U-Net, and the ResU-Net++ that we utilized. The empirical evidence reveals the superior performance of ResU-Net++ relative to its counterparts in computational proficiency, feature discernment capability, and inversion precision, thereby authenticating its applicability. Moreover, in the endeavor to predict salt mounds in the Nordkapp Basin situated in the Norwegian offshore Barents Sea, ResU-Net++ was able to delineate the boundaries of these mounds with heightened precision, reinforcing the pragmatic viability of this network model.