An Underwater Crack Detection System Combining New Underwater Image-Processing Technology and an Improved YOLOv9...

Area of Science:

• Civil Engineering
• Computer Vision
• Image Processing

Background:

• Underwater crack detection is crucial for infrastructure integrity but hindered by image quality challenges.
• Existing deep learning methods require extensive, hard-to-obtain underwater crack datasets.
• Hazardous underwater environments complicate data acquisition for crack monitoring.

Purpose of the Study:

• To develop an effective method for underwater crack detection by enhancing image quality.
• To adapt deep learning models for improved performance on processed underwater crack images.
• To provide a robust solution for detecting cracks in underwater dam structures.

Main Methods:

• A novel white balance and bilateral filtering denoising method was employed for underwater image enhancement.
• Underwater crack images were transformed into high-quality, above-water equivalents preserving original features.
• An improved YOLOv9-OREPA model was utilized for crack detection on the enhanced images.

Main Results:

• The proposed image processing technique significantly improved image quality metrics compared to existing methods.
• The enhanced images facilitated superior performance of the improved YOLOv9-OREPA model.
• Experimental results validated the effectiveness of the integrated approach for underwater crack detection.

Conclusions:

• The developed method offers a novel approach to transform challenging underwater crack images into usable above-water formats.
• This technique enhances the feasibility and accuracy of deep learning-based crack detection in underwater infrastructure.
• The study successfully demonstrates a practical solution for detecting underwater cracks in dams.