The Research Of Image Processing And Recognition Technology For X-ray Tube Weld Defection

Weld defect detection is a key link to ensure the quality of welding, with the rapidly development of industry and urgent demand, based on computer image processing and automatic identification technology of weld defect detection problem has been widely research. Among them, the X-ray detection has the imaging speed,real-time detection, low cost advantage, thus has attracted much attention of people.However, because of the imaging methods, and the influence of casting material and other objective factors, X-ray image noise background, low contrast and brightness uneven, weld edge blur, which makes use of computer to weld defects automatic detection accuracy is not very ideal. Automatic weld defects detection belongs to the field of computer image processing and recognition, which is a challenging task, it combines with signal processing, image processing and analysis, pattern recognition,computer vision, and other fields of cutting-edge technology, has great application value,though it is not a mature detection method yet.This paper first analyzes the domestic and foreign research present situation, the automatic detection of weld defect is pointed out that the current difficulties exist by weld defects automatic detection and related methods of image processing and identification technology, and puts forward an image segmentation based on scale product and depth of the defects of neural network classification based on sparse since the coding algorithm. The algorithm can be contained more noise and low contrast,weld edge blur the X-ray images of weld edge detection precision and segment the weld area. In the view of the current automatic weld defects detection field widespread recognition accuracy and the contradiction between the defect pictures miss rate, based on the depth of the neural network powerful ability of classification and recognition,puts forward a pair of complementary since sparse coding depth of neural network,enables under the lower miss rate get higher accuracy, namely in the practical application of judgment for zero defect image was higher and there is no defect images contained in the defect images the ratio of the smaller, so as to push the algorithm to the level of practical application. Algorithm is proposed in this paper is divided into two parts:(1) in the weld region segmentation, the algorithm through the different scales of weld image transverse gray-scale curve fitting, and by using both the scale of theproduct detect the edge of the weld, so as to extract the weld area;(2) in the classification and identification of weld defects, the algorithm proposed five sparse autoencoder deep networks, which four detect the sub-images with different sizes and different locations, and the last one detects the whole image. Through the analysis of those classification results, we can obtain the high accuracy under the lower miss rate.Finally, this paper performs an experiment about X-ray weld images, the experimental results show that the proposed algorithm is better than the other mainstream algorithm, can satisfy the practical application of the industry needs.