Crack detection for concrete bridges with imaged based deep learning.

Area of Science:

• Civil Engineering
• Computer Vision
• Artificial Intelligence

Background:

• Automated crack detection in concrete bridges is crucial for structural health monitoring.
• Existing image processing methods often lack precision and efficiency.

Purpose of the Study:

• To develop and evaluate a novel, combined deep learning approach for accurate and automatic bridge crack detection.
• To integrate this method into a portable device for practical field applications.

Main Methods:

• A dataset of concrete crack images was created through segmentation and preprocessing.
• Deep learning using the Single Shot Multibox Detector (SSD) algorithm was employed for initial crack identification.
• Sliding window technology and an eight neighborhood algorithm were used for detection and correction.

Main Results:

• The SSD algorithm achieved high precision and recall in detecting concrete cracks.
• The eight neighborhood algorithm effectively improved detection accuracy by reducing misjudged results.
• The developed algorithm demonstrated robust performance when integrated into a portable device.

Conclusions:

• The combined SSD and eight neighborhood algorithm offers a highly effective solution for automatic bridge crack detection.
• The portable system facilitates broader application in real-world concrete bridge inspection and maintenance.