An Electrical Capacitance Tomography Dataset for Image Reconstruction Benchmarking

Area of Science:

• Tomography
• Sensor Technology
• Machine Learning

Background:

• Electrical Capacitance Tomography (ECT) is a non-invasive imaging technique.
• Existing benchmarks often rely on simulations, lacking real-world complexities.
• Evaluating ECT algorithms requires comprehensive, experimentally acquired datasets.

Purpose of the Study:

• Introduce ECT-Bench, the first large-scale, experimentally acquired dataset for planar ECT.
• Provide a standardized benchmark for evaluating ECT imaging and dynamic tracking algorithms.
• Facilitate the development of robust ECT algorithms for real-world applications.

Main Methods:

• Acquired 54,739 high-resolution mutual capacitance matrices using a custom 3x3 electrode array.
• Collected data via a robotic pipeline covering diverse spatial variables, material types, and object shapes.
• Included raw capacitance signals, positional metadata, ground-truth labels, and quantitative reconstructions.

Main Results:

• ECT-Bench captures genuine soft-field characteristics, dielectric variability, and hardware noise.
• Compared traditional physical solvers (Linear Back Projection, Landweber) against deep learning (CNNs).
• Demonstrated the dataset's utility in evaluating algorithm performance under realistic conditions.

Conclusions:

• ECT-Bench establishes a rigorous foundation for advancing ECT technology.
• The dataset bridges the gap between numerical simulations and practical touchless sensing.
• Openly releasing fabrication files, acquisition codes, and baselines promotes reproducible research.