How Deep Learning Can Help Solving Geophysical Inverse Problems

Abstract This brief summarizes some of the main results I obtained during my Ph.D. studies at Politecnico di Milano, under supervision Professor Stefano Tubaro. The thesis provides contributions to understanding advantages, and limitations, data-driven deep learning approaches geophysical inverse problems, with a special focus on Convolutional Neural Networks (CNNs). Exploration Geophysics aims estimating accurate physical properties Earth subsurface from seismic data acquired close surface. Seismic show great variety statistically relevant independent patterns. devise Deep Learning methods solve several tasks by such First, generative networks as post-processing operator for refining reflectivity images. When trained pure image datasets, these suffer lack knowledge. Then, different approach named Priors, which are CNNs that precondition problem. In particular, develop scheme interpolate data. Finally, leverage features extraction ability buried landmine detection Ground Penetrating Radar (GPR) acquisitions. While presented effective compared state art, improvements can be achieved integrating algorithms within general problems theory through a-priori information derived domain