In response to the lack of equipment for targeted sowing of rootstocks for melons, as well as issues such as low efficiency in manual sowing and poor seedling quality, a method based on machine vision for identifying the geometry characteristics of white-seeded pumpkin seeds is proposed. Firstly, Jingxin No.2 rootstocks pumpkin seeds are selected as the research object, and industrial cameras are used to capture the original images of the seeds. The grayscale value, profile and morphological features of the seeds are obtained through grayscale transformation, mean filtering, and dynamic threshold processing. The best ellipse is fitted based on the direction and aspect ratio of the contour area, and the geometric center coordinates and long and short axis data of the seeds are obtained. The direction of the bud spots is determined by the maximum distance between the contour endpoints along the long axis and the geometric center point. Then, using trigonometric functions with the geometric center point as the reference, calculate the location of the bud spots, followed by experimental verification. The results of the experiment show that this method can effectively identify the position and angle information of the germination point, with an identification accuracy of 98.85%, an average angular deviation of 1.53°, and an average recognition time of 0.092 seconds for 25 seeds.