Automatic Parameter Selection of Image Reconstruction Algorithms for Planar Array Capacitive Imaging

• Published in: IEEE sensors journal 2018-08, Vol.18 (15), p.6263-6272
• Main Authors: Tholin-Chittenden, Carl, Abascal, Juan Felipe Perez-Juste, Soleimani, Manuchehr
• Format: Article
• Language: English
• Subjects: Capacitance; Computer Science; Electrical capacitance tomography; Electrodes; Engineering Sciences; Image Processing; Image reconstruction; landmine; Landmine detection; Permittivity; reconstruction; Sensors; Signal and Image processing; total-variation
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2018.2844549

Landmines are often be made out of plastic with almost no metallic components which makes detection difficult. A plausible solution is to detect superficial buried plastic objects using planar array electrical capacitance tomography (ECT). Distance detection is a big limiting factor of planar array ECT. Given the ill-posedness and loss of sensitivity with depth, regularization, and optimal selection of reconstruction parameters are required for detection. In this paper, we propose an "automatic parameter selection" (APS) method for image reconstruction algorithms that selects optimal parameters based on the input data based on a three step process. The aim of the first two steps is to provide an approximate estimate of the parameters so that future reconstructions can be performed quickly in step 3. To optimize the reconstruction parameters the APS method uses the following metrics. Front surface distance detection (FSDD) is a method of determining an accurate distance measurement from sensor head to object surface in low resolution image reconstructions using interpolation between voxels and Otsu thresholding. Cross-section reconstruction score (CSRS) is a simple binary image comparison method which calculates a ratio of expected image to reconstructed image. An initial set of capacitance data was taken for an object at various distances and used to train the APS method by finding the best reconstruction parameters for each distance. Then, another set of capacitance data was taken for a new object at different distances than before and reconstructed using the parameters selected by the APS method. The results of this showed that the APS method was able to select unique parameters for each reconstruction which produced accurate FSDDs and consistent CSRSs. This has taken away the need for an expert to manually select parameters for each reconstruction and sped up the process of reconstructions after training. The introduction of FSDD and CSRS is useful as they accurately describe how reconstructions were score and will allow future work to compare results effectively.