Radar forward modeling and intelligent identification of shallow subgrade defects

Abstract

Geological radar is the primary nondestructive testing method for evaluating shallow subgrade defects. However, the radar atlas contains a large amount of information, and the efficiency of manual data interpretation and processing is low. In this study, the characteristics of radar maps of different defects were analyzed via forward simulation using finite-difference time-domain technology. The instantaneous characteristic information of different defect maps was integrated using map post-processing technology to improve recognition and translation accuracy. Finally, the convolution neural network algorithm was used to conduct data recognition to achieve the intelligent recognition of subgrade defects with an average detection accuracy of 73.93 % based on the radar subgrade defect atlas dataset, and the results were practically verified. The results show that the developed approach can accurately distinguish subgrade shallow defect information in the radar atlas. This approach is useful for accurate and efficient identification of latent highway defects.