Abstract: In order to get spatial information of planting in time, protect and utilize Erigeron breviscapus, the fuzzy inter-ridge boundary and the difficulty in obtaining training data set of fine markers were solved. An unmanned aerial vehicle remote sensing planting information extraction method for Erigeron breviscapus based on the combination of RGB band maximum difference method and weakly supervised semantic segmentation was proposed. Firstly, Erigeron breviscapus was labeled at border level in order to make weakly labeled data set to reduce labeling time cost. Then, a lightweight U-Net network was used for weakly supervised semantic segmentation of the weakly labeled data set to achieve rough extraction of Erigeron breviscapus. Finally, the RGB band maximum difference method was used to remove the non-Erigeron breviscapus in the rough extraction results to achieve the fine extraction of Erigeron breviscapus growing area. The experimental results showed that the proposed method in IoU was 90.55%, 90.74% and 86.63%, respectively, in three selected Erigeron breviscapus scenes, and the accuracy was higher than object-oriented classification method and maximum likelihood method. The effectiveness of the method was verified by ablation experiments.