Research On Registration Method To Detect Detects In Ultrasonic TOFD Image

The ultrasonic time of light diffraction (TOFD) technique has high detection precision, robustness and low error rate at location and identification of defect size. In the detection of weld, TOFD is good at measuring and monitoring all kinds of detects. Thus TOFD has been widely applied to nondestructive defection and well developed.In this thesis, the weld testing technology and image processing by TOFD are studied. At the first, this paper introduces the TOFD technique principle, the image display and mathematical model. Then, a series of TOFD original images were obtained by TOFD experiment system, using the original images for image registration.This thesis introduces the principle of image registration and three common image registration method, Analyzing the advantages and disadvantages of the three methods and application of image types. Then through the study of matching algorithm, using a phase correlation matching algorithm, makes TOFD registration accuracy to sub-pixel level. Analyzing the influence of the actual situation on TOFD image, and evaluating the results on detail. Through experiment identifying the capacities of the algorithm in reducing ultrasonic TOFD image noise.In addition, the thesis details the characteristics of the SIFT algorithm and implementation process. And using SIFT method to register ultrasonic TOFD image. When use traditional SIFT algorithm to register, the fixed threshold cannot be applied to all images, so this paper used adaptive threshold method to detect feature points. Optimizing feature points, and eliminating false match points by RANSAC. It simplified the calculation and improved the accuracy and robustness of SIFT image match algorithm. Approximating calculation repetition rate is reliable and can be measured by RANSAC through experiment, and through the repetition matching the degree of TOFD image registration. The higher repetition rate, greater the degree of matching, but also that more reasonable threshold parameter; on the contrary, it indicates matching the low level and threshold parameter are not optimal.Finally, introducing the characteristics and implementation of edge extraction, applying canny edge extraction for image after registering by SIFT. Selecting a different threshold to adjust the number of detects and ensure every defect to be detected. Experiments show that the method can identify detect better, and obtain a series of parameter about detect, including curvature, geometric center position, detect size, the pixel size etc.