| 利用样本生成方法进行机载多光谱LiDAR数据深度学习分类 |
| |
| 引用本文： | 赵沛冉,管海燕,李迪龙,景庄伟,于永涛.利用样本生成方法进行机载多光谱LiDAR数据深度学习分类[J].测绘通报,2021,0(12):16-21. |
| |
| 作者姓名： | 赵沛冉 管海燕 李迪龙 景庄伟 于永涛 |
| |
| 作者单位： | 1. 南京信息工程大学遥感与测绘工程学院, 江苏 南京 210044;2. 武汉大学测绘遥感信息工程国家重点实验室, 湖北 武汉 430079;3. 上海航天电子技术研究所, 上海 201109;4. 淮阴工学院计算机与软件工程学院, 江苏 淮安 223003 |
| |
| 基金项目： | 国家自然科学基金（41971414；62076107）；福建省自然科学基金（2021J05059） |
| |
| 摘 要： | 机载多光谱LiDAR系统能够快速、准确地获取地物的空间几何和光谱信息,为地物覆盖分类和目标识别提供新的数据源。近年来,基于三维点云的深度学习算法取得了一系列突破性进展,然而直接将不规则的原始点云数据输入深度学习模型进行基于点的分类存在一定的困难。本文提出了一种基于FPS-KNN的样本生成方法,用于基于深度学习的机载多光谱LiDAR数据分类。该方法首先对输入数据进行归一化处理;然后利用最远点采样方法(FPS)和K近邻法(KNN)在输入数据中生成一系列规则大小的训练样本数据集。通过机载多光谱LiDAR数据的试验表明,该方法所生成的样本不仅符合卷积神经网络所要求的输入数据形式,而且能够确保对输入场景的完整覆盖。
|
| 关 键 词： | 多光谱LiDAR 点云样本 深度学习 地物分类 样本尺度 |
| 收稿时间： | 2020-12-02 |
Deep learning classification of airborne multispectral LiDAR data using sample generation method |
| |
| ZHAO Peiran,GUAN Haiyan,LI Dilong,JING Zhuangwei,YU Yongtao.Deep learning classification of airborne multispectral LiDAR data using sample generation method[J].Bulletin of Surveying and Mapping,2021,0(12):16-21. |
| |
| Authors: | ZHAO Peiran GUAN Haiyan LI Dilong JING Zhuangwei YU Yongtao |
| |
| Affiliation: | 1. School of Remote Sensing & Geomatics Engineering, Nanjing University of Information Science & Technology, Nanjing 210044, China;2. State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China;3. Shanghai Aerospace Electronic Institute, Shanghai 201109, China;4. Faculty of Computer and Software Engineering, Huaiyin Institute of Technology, Huaian 223003, China |
| |
| Abstract: | An airborne multispectral LiDAR system, which can quickly and accurately obtain the spatial geometry and spectral information of ground objects, provides a new data source for ground coverage classification and target recognition. In recent years, a series of breakthroughs have been achieved in deep learning algorithms based on 3D point cloud. However, it is difficult to directly input irregular original point cloud data into deep learning models for point-based classification. In this paper, a sample generation method based on PFS-KNN is proposed for deep learning based classification models using airborne multispectral LiDAR data. The method first normalizes the input data, and then farthest point sampling method and k-nearest neighbor method are used to generate a series of training sample data sets with regular size from the input data. Experiments with the airborne multi-spectral LiDAR data show that the samples generated by the proposed method not only meet the input data format required by the convolutional neural network, but also ensure the complete coverage of the input scene. |
| |
| Keywords: | multispectral LiDAR point cloud samples deep learning object classification sample size |
| 本文献已被 万方数据 等数据库收录！ |
| 点击此处可从《测绘通报》浏览原始摘要信息 |
|
点击此处可从《测绘通报》下载全文 |
|
