In the process of hybrid rice seed production，the staggered transplanting of male parent seedlings, as one of the crucial strategies to ensure the success of seed production, poses stringent requirements on time sensitivity and spatial accuracy. The widespread application of visual navigation brought unprecedented potential to this delicate operation process. However, challenges arise from the morphological differences of female parent seedlings at various seedlings ages, instances of missing seedlings within rows, and poor row linearity. To address these issues, an efficient and precise real-time guideline extraction method was proposed and validated through comprehensive experimentation. Firstly, a staggered transplanting dataset was created, incorporating different seedling ages to meet the needs of various rice varieties. Utilizing this dataset, the BiSeNet V2（a dual-branch segmentation network）was trained to extract the female parent row masks. The distance transformation of pixels within these masks was then used to extract the crop row centerlines, accurately representing the row positions. The nearest left and right row centerlines to the male parent area were extracted by using a segmented filtering method. The feature points of these centerlines were paired by using a rotational scanning method, and the midpoints of the paired feature points were used as the navigation line feature points. Finally，B-spline curves were employed to fit these guideline feature points, forming the final transplanting guideline. Semantic segmentation experiments demonstrated that the BiSeNet V2 achieved an average pixel accuracy, mean intersection over union（mIoU）, and inference speed of 88.73%, 57.47%, and 143.32 frames per second（f/s）, respectively. Guideline extraction experiments showed an average deviation of 4.66 pixels, a standard deviation of 2.73 pixels, and an extraction speed of 12.52 f/s. Field experiments further verified the effectiveness of the proposed method, showing an average deviation of 64.93 mm between the automatic navigation transplanting path and the manually marked optimal path, with a standard deviation of 51.96 mm and over 80% of positioning points having a deviation of less than 83.26 mm. In summary, the proposed guideline extraction method for male parent transplanting in hybrid rice seed production significantly enhanced the real-time, accuracy, and robustness of guideline extraction. This was achieved through the comprehensive preparation of the dataset, efficient segmentation of male parent rows, accurate extraction of crop row centerlines, correct pairing of feature points, and precise fitting of B-spline curves. The research result can provide a significant reference for the automatic navigation of male parent transplanting in hybrid rice seed production.