基于栈式自编码的倾斜摄影测量点云多层级分类方法  

Hierachical classification of oblique photogrammetric point clouds based on stacked Autoencoder

作　　者：何雪 邹峥嵘[1] 张云生[1] 陈斯飏 HE Xue;ZOU Zhengrou;ZHANG Yunsheng;CHEN Siyang(School of Geoseiences and Info-Physics,Central South University,Changsha 410083,China)

机构地区：[1]中南大学地球科学与信息物理学院,湖南长沙410083

出　　处：《黑龙江工程学院学报》2018年第3期1-6,共6页Journal of Heilongjiang Institute of Technology

基　　金：国家重点研发计划资助项目(2016YFC0803108);国家自然科学基金资助项目(41201472)

摘　　要：提出一种基于栈式自编码的倾斜摄影测量点云多层级分类方法。该方法首先利用栈式自编码对邻域内点云特征进行降维;然后将这些深度特征输入到随机森林分类器中进行训练获得分类器;最后结合语义信息对RF分类器分类的结果进行优化,得到最终的分类结果。选取2组典型数据进行实验,总体分类精度分别达到91.90%和88.57%,比基于K-邻域所有特征的分类方法提高4.98%和9.93%,比基于K-邻域均值特征的分类方法提高6.52%和7.44%。A hierarchical oblique photogrammetric point clouds classification method based on stacked autoencoder is proposed in this paper.The method firstly uses a stacked autoencoder to reduce the dimension of the features generated from the neighborhood of point clouds.Then these deep features are used to train a classifier based on random forest（RF）algorithm.At last,with the combination of the semantic information,the initial classification result derived from RF is optimized to obtain the final classification result.Two sets of typical data are employed for experiments.The overall classification accuracy is 91.90% and 88.57% respectively,which is 4.98% and 9.93% higher than the classification method based on all features of K-neighborhood,which is 6.52%and 7.44% higher than the classification method based on the average features of K-neighborhood.

关 键 词：栈式自编码 倾斜摄影测量点云 点云分类 随机森林 

分 类 号：P231[天文地球—摄影测量与遥感]