A new algorithm for detecting defects of sub-arc welding x-ray image based on compress sensor theory

An efficient X-ray radiography image analysis algorithm is developed for submerged-arc welding defects detection. The compress sensor theory is incorporated into the new algorithm, and the problem of defect detection is changed to a model recognition problem. The given X-ray image is represented by a linear combination of few model X-ray images. If a dictionary of model defect images and noise images are obtained, the coefficient vector can give important information for deciding the given image is defect or noise. Thus a sparse vector representation is sought by performing l0, l1 and l2 norm minimization. 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. JP2Tested with 800 x-ray radiography images obtained from a factory production line, the proposed algorithm achieves a sensitivity 99% and specificity 98%.