Extracting sparse crack features from correlated background in ground penetrating radar concrete imaging using robust principal component analysis technique

Crack detection is an important application for Ground penetrating radar (GPR) to examine the concrete road or building structure conditions. The layer of rebars or utility pipes that typically exist inside the concrete structure can generate stronger scattering than small concrete cracks to affect detection effectiveness. In GPR image, the signature patterns of regularly distributed rebars or pipes can be deemed as correlated background signals, while for the small size cracks, their image features are typically irregularly and sparsely distributed. To effectively detect the cracks in concrete structure, the robust principal component analysis algorithm is developed to characterize the rank and sparsity of GPR image. For performance evaluations, simulations are conducted with various configurations.