Defects Detection In X-ray Weld Seam Image

The moving small object detection in image is always a difficult problem in field of image processing, which applies in many fields, such as industrial detection and medical detection.The defects, such as blowholes and incomplete penetration, occasionally appear in the welding process. These defects can affect the quality and the security of products. Therefore, defects detection in welding seam is extremely important. Now, the on-line detection of defects in the weld is still done by human interpreter. However, this process is subjective, inconsistent, labor intensive and fatigue of interpreter. It is desirable to find an effective automatic defects detection method to assist human interpreter in evaluating the quality of weld and to make the on-line detection objective, standard and intelligent. Our research is based on this.We have studied the automatic defects detection in the weld seam and mainly done the following research:(1) There is much redundant background information for the defects detection in the image. Therefore we use an automatically abstracting method of weld area based on the auto-adapted threshold segmentation. This method can reduce the computation and increase the precision.(2) The SUSAN algorithm has good anti-noise ability, which can recognize the image edge very well. So we have studied a defects detection method based on SUSAN algorithm, which associated with the morphology operation. The results indicate that this method is effective.(3) Wavelet analysis method has a very good localization characteristic, which can focus on the arbitrary detail of the analyzed object. Therefore, we studied a method using wavelet decomposition to get the shape and position information of the defects. Then we use the wiener filter and morphology method to complete the detection.In order to confirm the validity of the two algorithms, we have detected the weld seam image with some defects, which obtained from the factory. The detected results are definitely accurate. It is proved that our algorithm is valid.