Abstract: In view of the phenomenon of false weld inspection images was very common in engineering, at present, it was mainly checked by manual reading, which was low efficiency and easy to miss the inspections, based on image processing a method was proposed to automatically locate, segment and compare the overlapping areas of the weld seams. An automatic detection system of weld seam flaw detection image overlap was established to detect whether the weld images were false. Based on the Radon transform, the overlap position marks were found, the overlap areas were cut from the weld images, and the overlap matching were performed using the SIFT algorithm to calculate the structural similarity of the two lap areas. Finally, according to the number of matching points of SIFT key points and structural similarity, they were judged whether the two overlap areas were matched. The experimental results show that the detection accuracy of automatic overlap detection method achieves 87.3%.

Key words: weld seam, nondestructive inspection, overlap detection, scale invariant feature transform（SIFT） algorithm, structural similarity

摘要： 针对焊缝探伤图像造假的现象在工程中比较普遍，而目前主要通过人工读片方式进行检查，检查效率低下、容易漏检。提出了一种基于图像处理的方法对焊缝的搭接区域进行自动定位、分割和比对，建立了焊缝探伤图像搭接自动检测系统，检测是否存在焊缝图像造假的现象。基于拉东变换找到搭接定位标记，从焊缝图像中切割出搭接区图像，采用尺度不变特征变换算法进行搭接匹配，计算两个搭接区图像的结构相似度。最后根据SIFT关键点的匹配点数和结构相似度判断两个搭接区图像是否匹配。实验结果表明，该方法的焊缝图像搭接自动检测准确率达到 87.3%。

关键词: 焊缝, 无损探伤, 搭接检测, 尺度不变特征变换算法, 结构相似度