Automatic weld recognition and extraction from real-time X-ray images using quadratic curve fitting and multi-order differences analysis of intensity profile

This paper presents a fast and adaptive method for weld extraction. Firstly, a fast algorithm was developed to recognise and alarm welds with large surface defects such as burn-through. Then, the upper and lower boundary ranges of the weld in the middle column of the image were estimated based on least-square quadratic curve fitting and a deviation threshold. The weld boundary ranges in the other columns were estimated by the weld edges of the adjacent detected column. Finally, the first-order and second-order differences of the intensity profiles in the estimated boundary range were analysed to judge whether there was a defect near or on the weld edge, and different approaches based on second-order differences analysis were subsequently proposed to extract the weld. Results show that the proposed method is effective to detect and extract different types of weld, with or without defects, in a real-time X-ray weld image.