A Multi-Scale Attention Network for Uncertainty Analysis of Ground Penetrating Radar Modeling

A multi-scale attention-based model (MSAM) is proposed as a surrogate for uncertainty analysis (UA) in ground penetrating radar (GPR) simulation. Instead thousand of full-wave simulations, the converts uncertain inputs to electric fields, and output effectively quantified. Global feature aggregation (GFA) module local affinity reconstruction (LAR) are presented improve representation capability by Affinity calculation under different receptive fields. In addition, new loss function accelerate convergence training data with wider range input disturbances. The UA result from Monte Carlo method (MCM) validates effectiveness model. comparison existing deep learning methods, can efficiently get higher quality predictions. Meanwhile, Sobol indices evaluated MSAM accord those MCM, which needs running simulation one times converge.