DAMVNet: Three-dimensional point cloud classification network based on dual attention mechanism and VLAD

Abstract

Aiming at the lack of effective use of contextual fine-grained local features in the existing deep learning-based 3D point cloud classification model, which leads to lower classification accuracy, a three-dimensional point cloud classification network based on dual attention mechanism and VLAD is proposed. Firstly, the local fine-grained features and global information of point cloud are mined by self-attention mechanism, and then the local geometric representation is learned by embedding graph attention mechanism in MLP layer. To take full advantage of the features, a multi-headed mechanism is used to aggregate different features from separate headers, and an effective key point descriptor is introduced to help identify the global geometry. Finally, the high-level semantic features of point clouds are obtained by locally aggregating vector VLAD layers. The experimental results show that the model achieves 92.45% accuracy on Mode1Net40 dataset.