Abstract: To solve the problem that the recognition accuracy of apple-picking robots is unstable under varying illumination conditions, a novel method based on illumination invariant graph was proposed for apple recognition. Firstly, the algorithm of medium filtering was used for the pre-process of apple image. Then the illumination invariant graph was calculated to eliminate the influence of illumination change. Finally, Ostu method was used for recognizing apple fruit. The experiment result illustrated that the stability of recognition accuracy of the method based on illumination invariant graph was three times of the method without illumination invariant graph under four kinds of illumination conditions, and the average recognition accuracy was up to 90.45%. The apple recognition method based on illumination invariant graph can overcome the negative effects brought by illumination change and improve the fruit recognition technology in outdoor environment.