Application Of An Improved Watershed Segmentation Algorithm In Weld Detection

Weld defects detection is an important technology in the manufacturing industry. The defects, such as blowholes, crack, slag inclusion and incomplete penetration, may appear in the welding process occasionally. The quality and the security of products can be affected by these defects. Therefore, defects detection in welding seam is extremely important. The defect detection methods on X-ray image are mostly based on threshold segmentation, edge detection or pattern recognition. Although we can acquire contain effects through these methods, it turns out to be relatively difficulty when the detection area is small or special resolution is low, and it is also hard to find different detects by using only one single method.In order to solve the problems that exist in the algorithm mentioned, a new method based on the watershed algorithm is proposed in this thesis. The arrangement is introduced as follow:Firstly, we review the history of weld defects detection technology and expose the disadvantages in detail by analyzing the classical methods in this field. After that, a feasible improving plan is presented.Secondly, by analyzing the property of random noise in X-ray image, we adopt an imaging filter method which is based on morphology, and this method decreases the noise effectively and raises the accuracy significantly.Thirdly, by discussing the method that inhibiting the over-segmentation in the watershed algorithm through experiment, we realize the watershed algorithm that combine gradient editing with process combination, and solve the problem existed in the traditional detection algorithm, and at the same time the automatic acquisition of the threshold are realized by using difference method during gradient editing and process combination.Finally, we prove the feasibility, adaptability and the accuracy of segmentation by computer simulation with the matlab6.5 tool box.