2.5-D Helicopter Transient Electromagnetic Data Inversion Method Based on SE-TransUNet

Abstract

Helicopter transient electromagnetic (HTEM) method exhibits significant advantages in geological exploration of complex terrain areas, yet its data inversion still faces challenges, such as low computational efficiency and poor inversion accuracy. Aiming at the problems of poor continuity of 1-D inversion and high computational cost of 3-D inversion, this letter proposes a 2.5-D HTEM inversion method based on the SE-TransUNet network. This method performs forward modeling on the subsurface resistivity model with infinite horizontal extension using a 3-D source to generate a dataset and uses an improved TransUNet architecture for inversion. The SE-TransUNet network combines the squeeze-and-excitation (SE) module and Transformer blocks on the basis of the U-Net neural network to enhance feature extraction and long-range dependency modeling, thereby improving the ability to reconstruct subsurface resistivity models. In experiments with synthetic and field data, compared with the Gauss–Newton method and the U-Net method, SE-TransUNet achieved better performance in the characterization of anomalous body boundaries, localization of anomalous bodies, and overall inversion accuracy, with an inversion time of only 5 s. The results verify the reliability and high efficiency of the proposed method for high-resolution HTEM inversion in practical geological exploration, providing an effective solution for fast and reliable HTEM inversion.