Rebar Radius Retrieval by Deconvolution and Convolutional Neural Network in Ground Penetrating Radar

The ground Penetrating Radar (GPR) is an effective method for construction quality inspection, typically of the 'thin' reinforcing rebars. However to the best of the author's knowledge, there is no widely-adopted method in detecting the rebars radius in concrete. Hereby, this paper proposes a novel machine learning framework to accomplish accurate and real-time estimation for the radius of the rebars, which can simultaneously estimate the burial depth of the rebars and the water content of the concrete. The proposed method mainly consists of two parts. First, a data pre-processing based on deconvolution is used to derive the reflectivity series of the rebars from a single A-scan. Then, a regression scheme based on one-dimensional convolutional neural network (CNN) uses the reflectivity series as input to accomplish the estimation. Simulation shows that the proposed method yields a high estimation accuracy of the radius.