Abstract: Aiming at nonlinear and unstable characteristics of seam tracking signal of magnetic controlled arc welding, the paper proposed a characteristic signal extraction method of arc seam tracking by combining Matching Pursuit and the nonparametric basis function. Firstly, adaptive adjustment of template signal approximate one characteristic component of the original tracking signal is performed, then the optimal estimate of signal characteristic component matching the Template Signal is figured out by the extraction method of characteristic waveform based on nonparametric basis function. Secondly, the optimal estimationis used to update signal margin according to the principle of Matching Pursuit and then the optimal estimation of other characteristic component in the remaining signal was obtained. Finally, the process is repeated again and again until the energy of signal margin reach the preset threshold value.The feature signal extraction experiment on seam tracking platform based on magnetric controlled Arc sensor was verified that the method can extract signal scanning groove and the feature signal corresponding to short circuit transfer.