AI-Based Graphical User Interface to Detect Debonded Pavement Sections from GPR B-Scan Data

In this paper, we present an AI-based Graphical user interface (GUI) devoted to B-scan data visualization, interpretation and classification of GPR data over debonded areas in the pavement structure. Two independent GUIs perform the processing of GPR data at two levels: gPRocessor and gprDetector. gPRocessor enables first an automatic preprocessing of the GPR data. It allows the operator to view, modify and update the data from any radar configuration in either time or frequency domains. The output B-scan processed data can be exported per user requirements. In the second stage, gprDetector implements the interlayer debonding detection on a scan-by-scan basis. Various supervised and unsupervised machine learning methods with suitable feature engineering techniques allow to classify A-scan data as either debonding or healthy zone. gprDetector includes a tool to label the data for defining the pseudo-ground-truth to be used for the performance assessment of the classification methods. Graphical results may be displayed at each processing stage to provide detailed information about the data. Both GUI tools were tested on the field data base in [1].