Research on weld edge detection based on improved canny algorithm

Abstract Aiming at the interference of light and scratches in the welding scene, the traditional weld edge detection accuracy is limited. A weld edge detection method based on the improved Canny algorithm is proposed. First, a Gaussian adaptive bilateral filter (GABF) is constructed to suppress the noise and retain the edge details. Then, a four-direction 3×3 operator template is introduced for gradient calculation. Then, the optimal threshold is selected by combining the adaptive threshold iteration algorithm and the maximum inter-class variance method. Finally, false edges shorter than the threshold are filtered out by calculating the number and length of edge point connectivity. The experimental results show that the improved algorithm achieves more than 95% in structural similarity and significant improvement in connectivity, which realizes the effective recognition of weld seams.