Abstract： In the automatic welding process of nuclear reactor vessels, the joints of the tubesphere intersections are usually complex. This paper presents a complete algorithm in detecting the intersecting line of the Jgroove based on image processing techniques and the support vector machine (SVM) classifier. Taking the combination of color moment feature and optimized gray level cooccurrence matrix (GLCM) feature as the feature vector, classification was made using the SVM classifier to acquire a binary image with the intersection area highlighted, while the image block sliding mechanism and voting mechanism were applied. The precise position of the intersecting line was detected by fitting a quadratic curve of the maximum outline. Experimental results show that the algorithm can endure a complex environment and can achieve realtime requirement. The accuracy of the SVM classifier is up to 95.6%, and the processing time of each image is less than 170 ms.