RECURSIVE LEAST SQUARES DICTIONARY LEARNING ALGORITHM FOR ELECTRICAL IMPEDANCE TOMOGRAPHY

• Published in: Progress in electromagnetics research C Pier C Vol. 97; pp. 151 - 162
• Main Authors: Li, Xiuyan, Zhang, Jingwan, Wang, Jianming, Wang, Qi, Duan, Xiaojie
• Format: Journal Article
• Language: English
• Published: Electromagnetics Academy 01.09.2019
• Subjects: Algorithms; Data mining; Image processing; Machine learning; Methods; Tomography; China
• ISSN: 1937-8718; 1937-8718
• DOI: 10.2528/PIERC19081001

Electrical impedance tomography (EIT) is a technique for reconstructing conductivity distribution by injecting currents at the boundary of a subject and measuring the resulting changes in voltage. Sparse reconstruction can effectively reduce the noise and artifacts of reconstructed images and maintain edge information. The effective selection of sparse dictionary is the key to accurate sparse reconstruction. The EIT image can be efficiently reconstructed with adaptive dictionary learning, which is an iterative reconstruction algorithm by alternating the process of image reconstruction and dictionary learning. However, image accuracy and convergence rate depend on the initial dictionary, which was not given full consideration in previous studies. This leads to the low accuracy of image reconstruction model. In this paper, Recursive Least Squares Dictionary Learning Algorithm (RLS-DLA) is used to learn the initial dictionary for dictionary learning of sparse EIT reconstruction. Both simulated and experimental results indicate that the improved dictionary learning method not only improves the quality of reconstruction but also accelerates the convergence.