In order to enhance the modeling capability of global features in deep learning-based 3D point cloud classification models and improve their generalization performance, a point cloud classification model based on the fusion of graph neural network and attention mechanism is proposed on the basis of PointNet. Firstly, the extracted features are used to make the model pay more attention to the global context information, suppress the noise information, reduce the redundant parameters, and enhance the modelling ability of the global features by increasing the channel attention module and the spatial attention module, respectively. Secondly, different K-values nearest neighbor searches are performed within multiple scales of sphere radius to construct the input features for encoding, which not only reduces the scale of the graph and training overhead but also enables the model to learn features at different levels. Finally, neighborhood information is aggregated and node features are updated through graph convolutional neural networks. The output features of different graph convolutional neural network layers are summed up to fuse multi-level features and improve classification accuracy. The proposed model is trained and tested on the public dataset ModelNet40, achieving an overall classification accuracy of 88.6%, which outperforms the commonly used 3DShapeNets, VoxNet, ECC, and PointNet models, demonstrating its superiority in point cloud classification.