基于UNet++和注意力机制的点云边缘检测算法

Edges in a point cloud are important features to represent the contour information of the three-dimensional target. To effectively extract the edge of the point cloud, an edge detection algorithm for point cloud based on UNet++ and attention mechanism is proposed. The model adopts the UNet++ network structure to solve the problem that feature fusion of existing models have limitation. The spatial attention module is added to the model to dynamically adjust the weight of neighborhood points around the center point in the process of down-sampling feature extraction. In addition, channel attention module is added to the model to make the model pay more attention to the important channels. Finally, the weighted cross entropy loss is used to mitigate the impact of the imbalance in the number of edge points and non-edge points. The model in this paper is trained and evaluated on the ABC dataset. Compared with other edge detection algorithms based on deep learning, it is proved that the method in this paper can obtain more accurate edges and has stronger robustness.