The 3D point clouds obtained from 3D scanner and multi-view data lack semantic information，leading to difficulties in discriminating the plant organ parts from the point clouds when the number of plant point clouds is large，or when different organs of the plant are similarly colored or obscured.To deal with the problem，this article proposes a three-dimensional semantic modeling method for soybean leaves embedded with a two-dimensional semantic prior.The semantic segmentation of soybean leaves based on Mask R-CNN was conducted.The three-dimensional reconstruction，fusion and learning of the segmentation results and multi-view data were performed to transfer the semantic information of leave from 2D semantics to 3D point clouds and obtain the point cloud semantic information of plant leaf.The 3D semantic model of plant leaves was established.The model was validated through multiple sets of potted soybean plant experiments.The length and width of leaf were extracted and compared with the manual measurement data.Results showed that the mean square error of the length and width of leaf was 2.53 and 1.52 mm，with the determination coefficients of 0.97 and 0.89，respectively.It is indicated that the proposed method can conveniently and accurately construct the 3D semantic model of plant leaves.