A Robust Hyperbola Recognition Model with Fitting-Errors-Based Eliminating in GPR B-scan Image

‍ ‍As a nondestructive tool， ground-penetrating radar （GPR） has been widely used for the investigation of the subsurface， but it is challenging to automatically extract information from GPR B-scan images. In this paper， a robust integrated model for automatically recognizing and fitting the hyperbolae from GPR B-scan images is proposed， which can eliminate non-hyperbolic clusters. Firstly， the preprocessing method which consists of the mean subtraction operation， the adaptive thresholding algorithm based on gradient， and the opening and closing operations is implemented. The mean subtraction operation is utilized to suppress clutter and noise. And the adaptive thresholding algorithm based on gradient could transform the B-scan image to the binary image. Then the opening and closing operations remove discrete noise points. Next， point clusters with downward-opening are identified by open-scan-clustering algorithm （OSCA）. After that， these point clusters are directly fitted by hyperbola fitting algorithm based on algebraic distance. Finally， based on the fitting results of these point clusters， the fitting-errors-based eliminating （FEE） method removes downward-opening point clusters without complete hyperbolic feature， thus all hyperbolic point clusters in the B-scan image could be recognized and fitted. This integrated model consisting of methods above can automatically and robustly extract information from GPR B-scan images. The experiments on synthetic and real datasets indicate the effectiveness of the proposed integrated model.