Interpretable Deep Learning Paradigm for Airborne Transient Electromagnetic Inversion

Propose an interpretable deep learning inversion paradigm that unifies the denoising and inversion of real-world data, simplifying the entire data processing workflow and enhancing processing efficiency. The network decouples noisy data into noise and signal factors, then performs inversion using the signal factors. Physical information is utilized as guidance, and the entire data processing is completed based on the signal factors, making the network’s results more reliable and interpretable. Inversion results on real-world data indicate that our model can consistently and accurately obtain the underground electrical structure quickly, without requiring manual preprocessing of the data.