There are many outliers in the point cloud data obtained by 3D scanner, such as noise, missing points and external points. The work aims to propose a strong robust 3D point cloud data classification method based on ensemble learning to improve the classification accuracy of 3D point cloud data in the presence of these outliers. An idea of ensemble learning based on maximum voting method was proposed, which integrated the classification results of two deep neural networks to improve the generalization and accuracy of the network; global feature enhancement and central loss function were used to optimize the neural network structure to improve the classification accuracy and robustness. The proposed method shortened the training time to 30 epochs, and improved the classification accuracy effectively in the case of noise, missing points and external points. The proposed el-3d algorithm has better robustness than the current point cloud classification methods in the case of noise, missing points and external points.