Combining the YOLOv4 Deep Learning Model with UAV Imagery Processing Technology in the Extraction and Quantization of...

Area of Science:

• Civil Engineering
• Computer Vision
• Structural Health Monitoring

Background:

• Bridges face risks from natural disasters, necessitating regular inspections.
• Traditional crack assessment is challenging due to accessibility, lighting, and complex backgrounds.
• Current methods often lack quantitative data and are labor-intensive.

Purpose of the Study:

• To develop an automated system for detecting and measuring bridge cracks using UAVs and deep learning.
• To overcome limitations of manual inspection methods for concrete structures.
• To provide objective and quantitative data for bridge integrity assessments.

Main Methods:

• Utilized a UAV-mounted camera for image acquisition of bridge surfaces.
• Employed a YOLOv4 deep learning model for object detection of cracks.
• Applied image processing techniques including grayscale conversion, local thresholding, and edge detection (Canny, morphological).
• Implemented scale methods (planar marker, total station) for accurate crack size measurement.

Main Results:

• The YOLOv4 model achieved 92% accuracy in crack identification.
• Crack width measurements were precise, with accuracy as fine as 0.22 mm.
• The system successfully enabled quantitative crack analysis from UAV imagery.

Conclusions:

• The proposed UAV-based deep learning approach offers an efficient and accurate solution for bridge crack inspection.
• This method enhances objectivity and quantifiability in structural health monitoring.
• It addresses the challenges of accessibility and environmental conditions in traditional bridge assessments.