A supervised deep feedforward neural network (SDFNN)-based image reconstruction algorithm for radio tomographic imaging

Abstract

Radio tomographic imaging (RTI) is an emerging imaging technique that utilizes the shadowing losses on links between multiple pairs of wireless nodes within the sensing area to estimate the attenuation of physical objects. By using an image reconstruction algorithm, the attenuations caused by the physical objects will be transformed into a tomographic image. The tomographic image provides information about the shape, size and position of an object. However, the process of reconstructing a tomographic image from the RSS measurements is an ill-posed inverse problem, meaning that a small number of errors or variations in measurements will lead to a significant impact on the image quality. The existing linear inverse solvers provide fast reconstruction, but the imaging results is non-satisfactory and inaccurate. On the other hand, the nonlinear inverse solvers produce a higher quality image but are computationally expensive. Studies of applying deep learning technique and neural networks in tomographic reconstructions to solve the ill-posed inverse problems have emerged in recent years. However, to the best of our knowledge, the studies conducted in solving the inverse problem of RTI system using deep learning technique are rare. Therefore, a supervised deep feedforward neural network (SDFNN)-based image reconstruction algorithm for the RTI system is explored in this study to determine the feasibility of deep learning technique in reconstructing a tomographic image using RSS measurements only.