Aiming at the edge-detection for a class of objects of irregular shape, such as sunflower seeds, but the traditional Canny algorithm has many shortcomings. This paper proposed three improvements: ① The adaptive high threshold method based on the gradient statistic difference was to overcome the problem that the traditional Canny algorithm might lose part valid edge information when the gray of goals and background in the image had a relatively wide change, which had higher signal-to-noise ratio; ② It proposed an algorithm of extending at the endpoints based on the gradient direction was made up for the defect of getting many interrupted edges after the edge link analysis of traditional Canny algorithm; ③ It proposed regional bounding algorithm based on the longest edge to remove the short edge of background for achieving the precise positioning of goals edges. The results demonstrated that improved Canny algorithm achieved good results in edge-detection of sunflower seeds in high complex background.