Real-time X-ray radiography for defect detection in submerged arc welding and segmentation using sparse signal representation

An efficient X-ray radiography image analysis algorithm is developed for defect detection in submerged arc welding and segmentation. The raw X-ray image is pre-processed to remove impulsive noise, enhance greyscale contrast and segment the regions of interest (ROI) where the welding images are located. A density-based spatial clustering method is then applied to detect and segment suspicious defect regions (SDRs) within the ROIs. These SDRs will then be normalised in size. An important assumption is that since the shapes of the SDRs are rather similar, a given SDR may be represented by a linear combination of very few model SDRs. In other words, if a dictionary of model defect images and noise images is provided, a given defect should then be better represented by the linear combination of true defect images rather than noisy images. A sparse vector representation is thus sought by performing l(0), l(1), and l(2) norm minimisation. Finally, the sparse representations of the defect part and noisy part are compared in the context of a maximum likelihood ratio test, which leads to the final classification. Tested with 400 X-ray radiographic images obtained from a factory production line, the proposed algorithm achieves a sensitivity of 99% (198/200) and a specificity of 98% (196/200). Compared to previously reported radiographic image analysis algorithms, the proposed algorithm is robust, efficient and readily applicable to real-world applications.