多分类器组合的LiDAR点云分类  

LiDAR Point Cloud Classification Based on Multi-classifier Combination

作　　者：鲁冬冬 邹进贵[1] LU Dongdong;ZOU Jingui(School of Geodesy and Geomatics,Wuhan University,Wuhan 430079,China)

机构地区：[1]武汉大学测绘学院,湖北武汉430079

出　　处：《测绘地理信息》2023年第3期36-40,共5页Journal of Geomatics

基　　金：国家自然科学基金(41871373)。

摘　　要：随着激光雷达技术在地形测绘、数字城市建设等领域的广泛应用,点云数据采集的精度和效率变高,若要进一步利用这些数据,还要对点云数据进行分类。因此,如何高效、高精度地对点云进行分类成为了目前的研究重点。机器学习不同分类器之间存在一定互补性,基于此,提出一种基于概率矩阵特征值的加权多分类器组合方法,利用矩阵的特征值来自适应调整分配权值,通过支持向量机和随机森林分类器组合进行点云分类。实验结果表明,该方法能有效结合两种分类器的优势,提高了分类的精度和稳定性。With the application of light detection and ranging(LiDAR)technology in terrain mapping,digital city construc⁃tion and other fields,the accuracy and efficiency of point cloud data collection have been improved.To further use these da⁃ta,the point cloud data should be classified.Therefore,how to classify point cloud data efficiently and accurately has be⁃come the current research focus.Based on the certain comple⁃mentarity between different machine learning classifiers,we propose a weighted multi-classifier combination method based on the eigenvalues of the probability matrix,which uses the ei⁃genvalues of the matrix to adaptively adjust the distribution weights,and classifies point cloud data through support vector machine and random forest.The experimental results show that this method can effectively combine the advantages of the two classifiers and improve the accuracy and stability of classi⁃fication.

关 键 词：点云分类 支持向量机 随机森林 分类器组合 

分 类 号：P237[天文地球—摄影测量与遥感]