A Virtual Channel Based Data Augmentation Method for Electrical Impedance Tomography
Jiahao Xu, Xiangyu Wang, Ningbo Yu, Jianda Han
Abstract
Image reconstruction in electrical impedance tomography (EIT) is ill-posed, manifested in the difficulty of estimating the dense conductivity distribution with less information. In the actual measurement, the restricted number of electrodes placed around the sensitive field is the main reason for the limited number of measured data. Aiming at this challenge, this article proposed a virtual channel based data augmentation (DA) method for EIT. By combining virtual channel technology and deep learning, we obtained a prior model for DA. A dataset containing 16000 samples was generated through EIT numerical simulations for model training. Then, the trained DA model was introduced before image reconstruction, and the distribution of conductivity changes was calculated with the sensitivity matrix-based algorithms. Both simulations and experiments proved that the proposed DA method can effectively improve the imaging quality without increasing physical electrodes. With the proposed method, the 8-electrode EIT system achieved comparable measurement performance with the 16-electrode EIT system.