Abstract: To overcome the problem that the existing supervision methods require a large number of training samples and the high cost of manual labeling, a supervoxel-based active relearning method is proposed. Firstly, feature extraction is performed on the point cloud. Secondly, training samples are selected through the active learning method with supervoxel constraint and manually labeled. Finally, the relearning method is used for post-processing, and the classification results are continuously optimized by iterative calculation of category statistical features. The experimental results indicate that compared to using all training samples, the proposed method can achieve the same overall accuracy using less than 6% of all samples, which greatly reduces the cost of manual labeling. And the relearning algorithm also improves the classification accuracy.

Key words: point cloud classification, supervoxel, active learning, relearning