Spatial resolution-enhanced electrical capacitance tomography using 32-electrode sensor and neural network training based on synthetic data

• Image spatial resolution improved using a 32-electrode ECT system with a high SNR. • First ML-based image reconstruction using real data from a 32-electrode ECT sensor. • The required SNR level for 32-electrode ECT measurements was determined. • ECT spatial resolution assessment using MTF. The low spatial resolution of images limits the broader application of electrical capacitance tomography. Image resolution could be improved by using more electrodes in the sensor to increase spatial sampling. However, increasing the number of electrodes at the expense of their area reduces inter-electrode capacitance, which in turn deteriorates the signal-to-noise ratio and limits the achievable improvement. This paper presents an improvement of electrical capacitance tomography spatial resolution using a 32-electrode single-ring sensor and machine learning-based image reconstruction. An artificial neural network architecture with fully connected layers was heuristically adjusted for the inverse problem. The network was trained on a synthetic dataset. The results of the experiments are provided for both numerically simulated and real measurement data. High signal-to-noise ratio front-end electronics were used to measure capacitance in the single-femtofarad range. Test objects developed for modulation transfer function evaluation were used to assess image spatial resolution. The study demonstrates that increasing the number of electrodes to 32 in a single-ring tomographic sensor improves image spatial resolution. The required signal-to-noise ratio for the measurement system to achieve increased spatial resolution was determined. The results also confirmed the effectiveness of the neural network trained on synthetic data when applied to real measurements.