DeepNIS: Deep Neural Network for Nonlinear Electromagnetic Inverse Scattering Article Swipe
YOU?
·
· 2018
· Open Access
·
· DOI: https://doi.org/10.1109/tap.2018.2885437
· OA: W2964269625
Nonlinear electromagnetic (EM) inverse scattering is a quantitative and\nsuper-resolution imaging technique, in which more realistic interactions\nbetween the internal structure of scene and EM wavefield are taken into account\nin the imaging procedure, in contrast to conventional tomography. However, it\nposes important challenges arising from its intrinsic strong nonlinearity,\nill-posedness, and expensive computation costs. To tackle these difficulties,\nwe, for the first time to our best knowledge, exploit a connection between the\ndeep neural network (DNN) architecture and the iterative method of nonlinear EM\ninverse scattering. This enables the development of a novel DNN-based\nmethodology for nonlinear EM inverse problems (termed here DeepNIS). The\nproposed DeepNIS consists of a cascade of multi-layer complexvalued residual\nconvolutional neural network (CNN) modules. We numerically and experimentally\ndemonstrate that the DeepNIS outperforms remarkably conventional nonlinear\ninverse scattering methods in terms of both the image quality and computational\ntime. We show that DeepNIS can learn a general model approximating the\nunderlying EM inverse scattering system. It is expected that the DeepNIS will\nserve as powerful tool in treating highly nonlinear EM inverse scattering\nproblems over different frequency bands, involving large-scale and\nhigh-contrast objects, which are extremely hard and impractical to solve using\nconventional inverse scattering methods.\n
