Measurement Image Reconstruction in Electrical Impedance Tomography through 1D-UNet

This paper presents a deep learning approach for image reconstruction in Electrical Impedance Tomography using a one-dimensional U-Net model. The model’s performance is evaluated against traditional methods such as the Total Variation and Gauss-Newton algorithms. Experimental results demonstrate that 1D-UNet consistently achieves superior reconstruction accuracy, particularly in noisy environments. In noise-free conditions, the model attains higher correlation coefficients and structural similarity values than conventional approaches, preserving fine details effectively. Under noisy conditions (30 dB and 60 dB), 1D-UNet maintains a significantly higher correlation and structural similarity, demonstrating its robustness. The strong generalization and adaptability of the proposed method underscore its potential for enhancing tomographic imaging applications in biomedical diagnostics, industrial process monitoring.