GPR Feature Enhancement of Asphalt Pavement Hidden Defects Using Computational-Efficient Image Processing Techniques.

Area of Science:

• Geophysics
• Civil Engineering
• Computer Vision

Background:

• Hyperbolic reflection features in ground-penetrating radar (GPR) data are crucial for identifying hidden asphalt pavement defects.
• Current deep learning methods struggle with raw GPR data, leading to inaccurate defect detection.
• Enhancing GPR data reflection features is vital for real-time, accurate defect analysis.

Purpose of the Study:

• To propose accessible image processing methods for GPR data preprocessing.
• To enhance hyperbolic reflection features for improved defect detection accuracy and speed.
• To evaluate the performance of preprocessing techniques on existing computer vision algorithms.

Main Methods:

• Applied image processing techniques like Sobel edge detection and histogram equalization to raw GPR data.
• Utilized standard image processing libraries for data preprocessing.
• Quantified performance using the regions of identified hyperbola signal-to-noise ratio (RIHSNR).
• Integrated preprocessed data with Faster R-CNN and CBAM-YOLOv8 models.

Main Results:

• Sobel edge detection and Otsu's thresholding significantly improved detection accuracy.
• mAP@0.5 increased from 0.65 to 0.85 for Faster R-CNN and 0.72 to 0.88 for CBAM-YOLOv8.
• Inference time decreased to 30 ms for Faster R-CNN and 25 ms for CBAM-YOLOv8.
• Enhanced hyperbolic feature detectability was confirmed via RIHSNR.

Conclusions:

• Simple image processing methods effectively enhance GPR data for pavement defect detection.
• Preprocessing raw GPR data improves the performance of deep learning models.
• The proposed methods offer a pathway for fast and accurate real-time GPR defect analysis.