Abstract: During the weld seam tracking process, intense arc light, smoke and spatter can severely disrupt the groove contour fringes that reflect the shape and position of the groove, causing abrupt jumps, breakage or even complete loss of these patterns. This poses significant challenges for subsequent feature point extraction, while the conventional image restoration methods remain computationally demanding and lack adaptability. Therefore, a one-dimensional convolutional neural network（1D-CNN） repair method was proposed based on an encoder-decoder structure to extract features and suppress noise through encoding, restore signal length through decoding, reconstruct the damaged area, and repair missing data, thereby repairing the damaged contour. The results show that compared with other methods, 1D-CNN has advantages in repair accuracy and computational efficiency,and can effectively repair the damaged groove contour data during welding. The average mean square error（MSE） of the repair is only 0.1757, and the single-frame contour repair time is no more than 0.0031 s, which fully meets the requirements of real-time weld seam tracking.