Research on GPR Response Recognition of Levee Hazards Based on Convolutional Neural Network

Abstract

As a key part of water conservancy and flood control, the safe service of levee is a matter of national importance. Since the construction of early levee showing the disease phenomena such as breakage, leakage and hollowness, it has seriously threatened the safety of people’s lives and properties, and became one of the important livelihood problems that need to be solved imminently. In this paper, the network model for automatic classification of levee hazards is constructed using convolutional neural network algorithm. The data set of the ground-penetrating radar responses of different hazards types required for training is constructed by forward modeling. Then, the network model is used to predict the hazards types from the actual measured data of a levee in Jiangxi Province. The results show that the good recognition accuracy and effect of the convolutional neural network model established in this paper.