Efficient Hyperbola Detection and Fitting Using Image Processing Techniques and Column-Connection Clustering

Ground Penetrating Radar (GPR) is a preferred non-destructive method used for the identification and localisation of subsurface targets. Hyperbolic signatures present in GPR B-Scans can provide valuable insights about the shape, material and the distance at which target objects are buried below the surface. This paper presents an innovative approach to enhance GPR data analysis, specifically focusing on hyperbola detection within GPR B-Scans. Preprocessing steps, such as dewow filtering, frequency filtering, and gain compensation, are used to improve GPR data quality. A comparative analysis is conducted on the performance of Canny, Sobel, and Scharr edge detectors in identifying hyperbolic signatures. By merging the Canny edge detection algorithm with the Column-Connection Clustering (C3) method, common limitations of conventional clustering, particularly their sensitivity to noise and outliers are addressed. Finally, the efficacy of the proposed method is evaluated by calculating the R-squared values of the fitted hyperbolas which was found to be > 0.93.