Research On Recognition Of Defects In X-ray Image Based On Wavelet Transform

Detecting welding line with X-ray is one of hot research areas in non-destructive testing. Judging the quality of X-ray film is an important part in X-ray detection. Traditionally, the work is done by judging worker, which makes the traditional testing method have many disadvantages, for instance, low in nicety and efficiency, different detection results between different workers and so on. With the development of computer technology, such as image processing and pattern recognition, it is an effective way to research and develop computer aided recognition system.The image preprocess is the base of the detection and recognition of defects in X-ray image. After analyzing the characteristics of X-ray image, effective image preprocess methods are studied in this paper. Polynomial warping and bilinear interpolations are applied respectively to calibrate the space position and the gray value of pixels. The experimental result shows the effectiveness of the methods. The multi-scale and multi-resolution trait of wavelet transform is utilized to process images. Combined with the soft threshold, the high frequency intensification method is used to enhance the defect's edge in X-ray image. This method not only reduces the noise effectively, but also intensifies the image detail and preserves the edge character of defects.Wavelet transform has become a good tool for edge detection due to its speciality of multi-scale and the ability to detect local abrupt upheaval. An improved image edge detection algorithm, mainly modifying the edge point judge criterion, is proposed. According to multi-scale analysis and local adaptive threshold, a wavelet transform is performed on the image by quadratic B-spline wavelet and Mallat algorithm to detect the edges of defects. The experimental results show that the aforementioned algorithm works better than traditional edge detectors, for it can detect multi-scale edge information and obtain single-pixel-wide defect edges.According to the computed feature parameters, fuzzy neural network is utilized for defect classification. The overall framework of X-ray image defects recognition system is designed. The function of the four modules and their design formula are introduced. Delphi and Matlab are used respectively to develop applications and process background computation. To accomplish the design and realization of defects recognition system, Mideva is applied to compile dynamic link library that is departed from Matlab environment to achieve mixed programming. In this way, the time for the system development is dramatically shortened. The practical application results show that the system designed can effectively recognize the defects in weld. For most of the defects, the detection results have achieved technical performance indicators.