An Integrated GPR B-Scan Preprocessing Model Based on Image Enhancement for Detecting Subsurface Pipes.

Area of Science:

• Geophysics
• Signal Processing
• Non-destructive Testing

Background:

• Ground-penetrating radar (GPR) is crucial for nondestructive subsurface pipe detection.
• Current automatic segmentation models struggle with diverse pipe sizes, subsurface conditions, and noise in GPR B-scans.
• Accurate hyperbola segmentation is essential for reliable GPR data interpretation.

Purpose of the Study:

• To develop an automatic preprocessing model for enhancing GPR B-scan interpretation.
• To improve the generalization capability of automatic hyperbola segmentation models.
• To address challenges posed by noise, varying pipe radii, and complex field conditions.

Main Methods:

• A novel automatic preprocessing model combining ground reflection removal algorithm (GRRA), data gravitational force enhancement (DGFE), and a global-local thresholding technique (DLTS).
• GRRA utilizes frequency and spatial filters to remove ground reflections.
• DGFE amplifies target hyperbolas, and DLTS segments them using dilation-based local thresholding.

Main Results:

• The proposed model effectively enhances GPR B-scan interpretation under challenging conditions.
• Demonstrated superior performance in segmenting diverse and small-scale hyperbolas amidst noise.
• Achieved significant cross-dataset generalization capabilities compared to state-of-the-art methods.

Conclusions:

• The developed automatic preprocessing model significantly improves GPR data interpretation for subsurface utility detection.
• The model offers enhanced robustness and generalization for hyperbola segmentation in complex GPR environments.
• This approach provides a more reliable tool for nondestructive subsurface pipe detection.