This paper introduces a novel approach for 3D Electromagnetic Tomography (EMT), designing a 3D EMT measurement system for volumetric conductivity reconstruction and defect detection, enabled by an innovative hybrid that combines a sensitivity-based algorithm with a Transformer–UNet model. Traditional non-iterative algorithms such as linear back projection (LBP), Tikhonov Regularization (TR), and Singular Value Decomposition (SVD) face limitations such as reduced accuracy in complex scenarios and susceptibility to noise. To overcome these issues, we propose a TR algorithm enhanced by bi-Laplacian regularization, significantly improving reconstruction quality. Extensive simulations and experimental validations demonstrate that the proposed method outperforms conventional algorithms, achieving higher quantitative gains. Furthermore, integrating deep learning techniques, specifically the proposed method, enables precise reconstruction by effectively combining raw electromagnetic signals with the prior reconstructed results based on the proposed reconstructed algorithm, thereby significantly improving robustness, accuracy, and detail preservation. Experimental verification confirms the practical efficacy and robustness of this hybrid model for industrial applications. This framework differs from prior purely data-driven reconstructions by explicitly coupling a sensitivity-based EMT solver with a Transformer–UNet trained with a supervised image-domain loss, where a physics-based pre-reconstruction is provided as an additional prior, enabling real-time monitoring with offline shape-accurate refinement. • A 3D EMT framework is proposed for conductivity and contour reconstruction. • A triple-regularized scheme improves volumetric imaging and defect depiction. • A dual-branch network enhances reconstruction accuracy and robustness.
She et al. (Mon,) studied this question.