Research on Point Clouds Target Recognition Method Based on Deep Learning and Ensemble Learning
In view of the problem that the performance of the current point cloud target recognition task is good while the im-provement is slow,and the single network model is not sufficient to extract features resulting in hidden loss of information,the re-search proposes a new point cloud target recognition method based on deep learning and ensemble learning.Based on the Stacking algorithm,the method uses Logistic regression to ensemble three types of typical point cloud target recognition deep neural net-works,which are PointNet based on neighborhood feature learning,RGCNN based on graph convolution,and PointCNN based on optimized χ-convolution.The experimental results show that the classification accuracy of the ensemble model on ModelNet40 datas-et reaches 93.7%,and the semantic segmentation mIoU on ShapeNet Part dataset reaches 87.6%,which is better than most existing deep learning methods.
deep learning3D point cloudstarget recognitionensemble learningStacking algorithm
国家科技期刊平台
NSTL
万方数据
