Weld Defects Detection And Recognition Technology Research Of X-ray Image

In industrial NDT domain, a variety of defects can bring a lot of unsafe accidents to real life in the weld.Currently, using artificial way of film quality evaluation is easy to exist subjective criteria inconsistent, low detection efficiency, complex operation and film materials is not easy to save, etc. Therefore, the research of automatic detection of weld defect method instead of artificial way is necessary.According to the above problem, this article mainly study of three aspects: the content of the weld defect pretreatment, weld defect segmentation and weld defects recognition.The specific as follows:(1) There are a large number of background regions in the X-ray image.Hence, the weld image is divided into three sub regions(the background area, the plate area and the weld area). An automatic segmentation method based on double Otsu threshold is used to improve the elapsed time of the defect detection.This method can get a better segmentation result.(2) Due to the image with low contrast and detection annihilation in weld. An image by calculating cumulative distribution values is used to determine gray stretching limit, adjust the image gray value, enhance the contrast of image, improve the display effect of the target area, improve target identification rate and achieved ideal effect.(3) Due to mathematical morphology open operation, can effectively eliminate the high frequency message of the image.If the size of the structural elements is constant.The processing results will result in the following two situations: First,when the defect size is larger than that of structural elements, high frequency part of defect regions can not effectively eliminate defect distortion; second, when the size of structure element, the processing speed will be slow. Therefore, a morphological filtering method which can change the size of the structure element with the defect size is designed.This method is very effective to induce the noise.(4)The SUSAN algorithm has good anti noise ability, This paper presents SUSAN template adaptive threshold selection method to improve the detection effect.(5) In order to mark the defect target, it is convenient to calculate the characteristic parameters of each defect. This paper makes use of the two scanning method to mark the defect target quickly. At the same time, in order to facilitate the calculation of characteristic parameters, the Moore boundary tracing algorithm is used to track the 8 connected coding of each defect.(6) The design of the weld defect classification depth of 4 of two binary tree classifier, and based on the values of the parameters in the literature, designed the recognition process reasonably. And the classification accuracy is verified by a large sample, achieved good classification results.In order to verify the rationality and real time detection method, I test the thirty two defects samples. The test results show that the detection method of experimental design can effectively on 31 defect classification recognition. For single image, the image size is 512 * 512, the number of defect about five X-ray weld defect image, the processing speed is about 0.3 second, can meet the requirements of defect detection and recognition.