An intelligent identification method to detect tunnel defects based on the multidimensional analysis of GPR reflections

Due to the rapid construction of tunnels in China, problems that are associated with both quality and safety have become apparent. Therefore, the control and treatment of various tunnel defects are gradually becoming a primary focus during both construction and operation of tunnels. Further, a ground penetrating radar (GPR), which is based on the ultra-high frequency pulse electromagnetic wave theory, provides advantages such as efficiency and convenience. Further, GPR has been extensively used to perform nondestructive detection of tunnel defects in order to ensure sufficient quality and safety. To improve the efficiency and reliability of the GPR detection process, a novel method that identified tunnel defects using the GPR images in an intelligent manner was proposes based on the multidimensional analysis of GPR reflections. Six typical identifying features of defect signals were initially extracted based on time domain, frequency-domain, and time-frequency domain analyses. Further, automatic identification of the horizontal distribution of the defect was obtained by searching for all the defect signals using a classification model constructed by a support vector machine, which was used for training the model with the typical features. Furthermore, by calculating the depth distribution of defects according to the first intrinsic mode function (IMF1) component envelope of the defect signals, intelligent identification of tunnel defects can be achieved. A comparison between the results of the intelligent and artificial identification mechanisms when applied to a tunnel backfill measured GPR data depicts that the intelligent method illustrates a strong ability to identify defects in GPR data. Further, only a few errors are produced:the identification rate and accuracy of test data are 100% and 78.6%, respectively, which satisfies the engineering application requirements. This method can be used to intelligently identify the defects in different types of GPR data in tunnel engineering. Furthermore, the results of this study can provide some hints about the design of intelligent identification algorithms that can be applied in other areas of GPR detection with various detection target types.