Automatic weld defect detection in real-time X-ray images based on support vector machine

Automatic weld defect detection based on real-time X-ray image plays a vital role in improving the automation level of radiographic inspection in industry. Most of the existing real-time automatic inspection technologies only use defect segmentation algorithms, which leads to the difficulty of reducing both the undetected rate and false alarm rate. In this paper, an effective method based on Support Vector Machine (SVM) is proposed to detect weld defect in real-time X-ray images. Firstly, all potential defects are segmented by background subtraction algorithm. Then three features including defect area, average grayscale difference to its surrounding district and grayscale standard deviation are extracted. Lastly, the extracted features are used as input to SVM classifier to distinguish non-defects from defects. Results show that the proposed automatic defect detection method can reduce the undetected rate and false alarm rate effectively in real-time X-ray images of weld.