An Image-to-image Target Reconstruction Network for Capacitively Coupled Electrical Resistance Tomography Based on Transfer Learning

This work develops an image-to-image target reconstruction network for capacitively coupled electrical resistance tomography. The novel target reconstruction network is constructed by two Unets to extract and fuse the multi-frequency features of the coarse images reconstructed by traditional image reconstruction algorithms, and incorporated with transfer learning to improve the generalization ability of the network. Experimental results show that the developed target reconstruction network is effective. The Unet-based image-to-image framework improves the target reconstruction quality with quantitatively higher image score. The introduced transfer learning strategy fills the gap between simulation data and experimental data, and further improves the performance of the network.