With the increasing demand for intelligent construction in science and technology, semantic segmentation technology has attracted extensive attention from scholars in the field of graphics and images. This technology provides effective decision support for target tracking, visual control, and other technologies. However, the operation efficiency and segmentation accuracy of the three-dimensional (3D) point cloud semantic segmentation model are bottlenecks to its development. A semantic segmentation network model of the 3D point cloud, called point cloud+graph convolution network (PCGCN) is proposed. PCGCN uses the EdgeConv network to extract local features and ResNet to enhance the transmission of features, fuse the local features of different scales, and participate in semantic segmentation of the 3D point cloud. In the process of deep learning, PCGCN solves the problem of the lack of local features and improves the segmentation effect. Furthermore, in the point cloud deep learning network, the introduction of ResNet improves the accuracy of semantic segmentation. Experiments are carried out using ShapeNet and S3DIS datasets. The experimental results show that the PCGCN accuracies are 85.1% on the ShapeNet dataset and 81.3% on the S3DIS dataset.