In order to find weld defects in time and improve the welding quality of workpieces, a laser vision image-guided mechanical weld recognition method is proposed. Analyze the structure of the laser vision imaging system, build the camera imaging model, and adjust the imaging position through coordinate transformation to avoid distortion; Through binary processing and expansion operation, the contour features of the pattern are retained to reduce the image noise; The gray distribution of the image is adjusted by contour pose correction and histogram equalization; Taking crack, incomplete penetration, incomplete fusion, porosity and spherical slag inclusion as the main manifestations of the weld, combined with the parameters of edge smoothness and filling degree, the weld characteristics are extracted; Establish the classification surface of support vector machine, input the samples to be recognized, input the feature types into support vector machine, determine the penalty coefficient, and output the recognition results after continuous training. Experiments show that the image obtained by the proposed method can contain more feature information and reduce the probability of error recognition.