Research on magnetic induction tomography method based on adaptive singular value decomposition algorithm

Magnetic induction tomography (MIT) is a non-contact, low-cost electromagnetic imaging technique with broad applications in industrial monitoring, biomedical imaging, and non-destructive inspection. However, existing MIT imaging algorithms suffer from slow reconstruction speed, limited imaging accuracy, and weak robustness to noise, which hinder their practical deployment. To overcome these limitations, this paper proposes an adaptive truncated singular value decomposition method based on the discrete Picard criterion, termed Picard-Driven TSVD (PTSVD). The method quantitatively analyzes ill-conditioned matrices via Picard curves, establishes an adaptive regularization parameter selection strategy, and integrates Tikhonov-Gaussian filtering to enhance noise suppression. Compared to traditional Tikhonov and TSVD methods, PTSVD reduces reliance on empirical parameters, avoids excessive image smoothing, and improves noise robustness. Simulation results show that under noise-free conditions, the image correlation coefficient increases by 28.8% and 75.9%, respectively. Even at a signal-to-noise ratio of 5 dB, the correlation coefficient remains at 0.63. Experimental validation on a self-built MIT platform further confirms the algorithm’s effectiveness and practicality. This study provides a high-accuracy, noise-resistant solution for real-time MIT detection, showing strong potential in applications such as molten metal flow monitoring, multiphase flow analysis, and subsurface target detection.