Aiming at the problems of underutilization of local feature information of three-dimensional (3D) point cloud and poor classification accuracy of existing 3D point cloud classification methods, a 3D point cloud classification method based on deep feature reinforcement is proposed in this paper. The method adopts a depth feature reinforcement-based 3D point cloud classification network to fuse the local feature information and global feature information of the 3D point cloud. And the proposed classification network adds two modules into the PointNet++ model, namely a point feature encoding module and a deep feature enhancement module. By feature encoding the 3D point cloud, the first module can extract local geometric feature information of the 3D point cloud. And by using an improved multilayer perceptron (MLP) to process the 3D point cloud, the second module can extract high-dimensional deep feature information of the 3D point cloud. By using these two modules, the proposed network can enhance the extraction abilities of the local features and the deep semantic feature simultaneously, and can enhance the generalization ability of the classification network as well. Experimental results on ModelNet40 dataset show that the proposed method has high accuracy in 3D point cloud classification, with an overall accuracy (oAcc) and a mean class accuracy (mAcc) of 93.4% and 90.9%, respectively.