Abstract: Groove signal recognition is a difficult point in automatic welding process. Based on this, the light path diagram was designed with the photosensitivity of PSD sensor and the corresponding light path model was established. Aiming at the nonlinear distortion of the PSD sensor to a certain extent, a corresponding improved model was established. According to the characteristics of the output displacement and time curve of the PSD sensor, a relevant algorithm was proposed to extract the corner information of the groove, so as to realize the real-time extraction of the diagonal points and the shape fitting of the groove. The results showed that the signal recognition system could accurately detect the change of groove position and defects of groove in real time, so as to avoid errors in the later welding process and affect the welding quality.