Interpretable deep learning paradigm for airborne transient electromagnetic inversion

Propose interpretable deep learning inversion paradigm integrates denoising and inversion of field data, streamlining the entire processing workflow while enhancing computational efficiency. The network explicitly decouples noisy input into noise factors and signal factors, subsequently employing the signal components for inversion. Guided by physical constraints throughout the processing chain based on these signal factors, the framework ensures both reliability and interpretability of network outputs. Field data inversion results demonstrate that our model achieves unified, accurate, and rapid reconstruction of subsurface electrical structures without requiring manual data preprocessing.