A region-based adaptive border peeling method for enhanced image reconstruction in capacitively coupled electrical resistance tomography

Electrical tomography (ET) offers a low-cost and useful tool for industrial multi-phase flow monitoring, though developing high-quality image reconstruction algorithms remains a major challenge. This work proposes a novel region-based adaptive border peeling (RABP) method to achieve enhanced image reconstruction in ET systems. The proposed method consists of three parts, including initial image reconstruction, pixel-level adaptive clustering by an improved border peeling (BP) algorithm, and morphological post-processing process. Firstly, an initial image of the flow distribution is reconstructed from impedance measurements acquired by the ET system using the linear back projection algorithm. Then, an improved adaptive BP algorithm is employed to automatically determine the reverse K-nearest neighbors (RKNN) of the pixels in the initial image through distance ratio jump truncation, and accurately identify the gas-phase targets within the liquid-phase background via connected component analysis (CCA) and local window statistics. Finally, morphological image post-processing operations guided by the forward model of ET, including convex hull and adaptive dilation/erosion, are applied to calibrate the shape deformation of gas-phase targets. With a 12-electrode capacitively coupled electrical resistance tomography (CCERT) system, simulation and experiments were conducted to validate the effectiveness of the proposed RABP method. Research results show that the RABP method consistently delivers high imaging performance across different two-phase distributions and outperforms other clustering-based reconstruction methods in image quality. In experiments, the average relative image error (RIE), structural similarity index measure (SSIM) and image correlation coefficient (ICC) of the images reconstructed by the proposed RABP method are 0.0535, 0.8280 and 0.8307, respectively.