A Hybrid Convolutional Feature Reconstruction Network for Electrical Resistance Tomography Image Reconstruction

Electrical resistance tomography (ERT) is a non-intrusive imaging method employed for capturing internal conductivity distributions of an object or medium by applying currents through boundary electrodes and collecting the resulting voltages. The primary challenge in ERT image reconstruction is caused by the inverse problem, and the traditional image reconstruction methods suffer to achieve high accuracy and resolution. To improve the image reconstruction accuracy, the Hybrid Convolutional Feature Reconstruction Network (HCFR-Net) is proposed. Fully connected layers are mostly used in CNN in which the size of inputs remain constant so they fail to capture spatial features from the data which causes less accurate reconstruction. To maintain important details in the data and enhance spatial features through the process of reconstruction, the up-sampling layer is introduced with FCN element to mitigate these issues successfully. The outcomes from simulation and experimental tests show that the proposed HCFR-Net algorithm has higher accuracy for image reconstruction.