Abstract: Fast and high-precision point cloud classification algorithms have wide applications, such as autonomous driving, 3D scene recognition, map reconstruction, and industrial robots. Deep learning requires huge amounts of labeled training data and does not consider the local information of 3D point cloud data in the network. Hence, iwe design a 3D point cloud framework based on the maximum classifier discrepancy domain adaptation method in order to resolve the issues of traditional 3D point cloud classification algorithms. For this, we first used the PointNet point cloud classification network as the basic framework of the network and then added a self-adaptive node module to the feature network to process the local features of the 3D point cloud. We then train the global features of the network by using the maximum classifier discrepancy in the domain adaptation method; this effectively alleviates the dependence on massive training data. The proposed method is tested on six migration combinations of the 3D point cloud datasets PointDA-10, and our results show that the classification accuracy of five combinations is better than that of the traditional point cloud classification algorithm. In addition, the proposed method can effectively improve classification accuracy when the amount of training data is reduced by 20% compared with the traditional methods.