Neural Networks applied on GPR data to create predictive GPR profiles
Predictive GPR Profile; Deep Convolutional Neural Networks; Style Transfer; Video Frame Interpolation.
This work proposes a new methodology for the creation of Predictive GPR Profiles, which are virtual radar images, built from deep Convolutive Neural Networks. Contents and styles are extracted from the original GPR profiles and combined with a loss function applying the technique: (i) Style Transfer, and characteristics of the GPR profiles obtained in the form of kernels involved with real data: (ii) Frame Interpolation. In this proposal, GPR profiles were acquired in outcrops analogous reservoirs encompassing different geological contexts and imaging depositional, deformational and karstic features.