Abstract: Aiming at the accuracy of X-ray weld image defect detection,this paper proposes to use the distance invariance and angle invariance of Log-Polar transformation to transform the position and shape of the defect into the translation of a simple two-dimensional plane of typical defect images, and solve the defects and suspected defects. area calibration problem. In addition, in order to improve the detection rate and recognition accuracy of defect identification, a defect identification based on sparse description is proposed, which uses three major knowledges: extracting typical samples from massive data, building non-parametric models, and solving sparse solutions based on the optimal direction method. system to identify the calibrated SDR. Experiments show that the recognition rate of the dictionary matrix obtained by limited sample training has reached more than 98.5%.