Attribute Recognition of Buried Pipes Based on Multi-Trace Phase Features Using K-means Clustering for GPR Data Interpretation

Accurate location and depth determination of underground pipes, especially the attribute recognition, are of great importance yet remake a challenging issue in municipal environments. Single-trace phase difference analysis remains a bottleneck due to its inherent and strong randomness in object identification. This paper developed a multi-trace phase difference analysis framework for ground penetrating radar (GPR) data based on K-means cluster analysis technique and the theory of region of interest (ROI), which could serve as a new criterion for successful pipe attribute recognition. After improving signal-to-noise ratio of GPR data by using the preprocessing techniques, the connected components algorithms (CCA) based on image segmentation and morphological operation is performed to delineate the ROI. The K-means cluster analysis technique is further employed to efficiently extract the multi-trace phase statistical features for comprehensively evaluating the attributes of ROI. We verify this proposed framework by simulated GPR signals, laboratory data and field datasets. Results demonstrate that the proposed method can not only facilitate the attribute recognition of pipes, but also reduce the interpretation ambiguity of the pipe material even in the field site environment. Specifically, if the phase difference of pipe turns out to be even multiples of p, the target can be automatically identified as metallic-category pipes, whereas odd multiples of p, point to non-metallic-category pipes with a lower permittivity than that of the background. This criterion presents promising applicability in subsurface pipeline identification and attributes recognition, especially in constructing a more appropriate initial model of GPR full waveform inversion for survey in pipes.