Welding defects extraction method by fusing saliency information of mid-level and underlying level images

Area of Science:

• Materials Science
• Non-Destructive Testing
• Image Processing

Background:

• X-ray non-destructive testing (NDT) is crucial for identifying weld defects.
• Challenges in X-ray weld images include low contrast, noise, and background variations.
• Existing methods struggle with low-contrast defect extraction.

Purpose of the Study:

• To develop a novel defect segmentation algorithm for X-ray weld images.
• To improve the accuracy and efficiency of weld defect detection.
• To address limitations of traditional methods in handling image quality issues.

Main Methods:

• Integration of underlying and mid-level image information.
• Utilizing a visual saliency model for initial feature extraction.
• Employing a Pulse Coupled Neural Network (PCNN) for mid-level saliency computation.
• Combining saliency maps using a pixel-minimum method for final segmentation.

Main Results:

• The proposed algorithm demonstrates high accuracy in defect segmentation.
• The method proves broadly applicable across various weld defect types.
• Rapid extraction of defects within the welding area is achieved.

Conclusions:

• The novel algorithm effectively overcomes challenges in X-ray weld defect detection.
• This approach offers a significant improvement over traditional methods.
• The technique is suitable for practical industrial applications in welding quality control.