Construction Method of Point Clouds' DEM Based on K-means Clustering and RBF Neural Network

被引：0

作者：

Zhao Q. ^{[1
,2
]}

Li P. ^{[1
,2
]}

Ma Y. ^{[3
]}

Tian W. ^{[4
]}

机构：

[1] College of Information Science and Technology, Shihezi University, Shihezi

[2] Division of National Remote Sensing Center, Xinjiang Production and Construction Corps, Shihezi

[3] Geospatial Information Engineering Research Center, Xinjiang Production and Construction Corps, Shihezi

[4] College of Mechanical and Electrical Engineering, Shihezi University, Shihezi

来源：

Nongye Jixie Xuebao/Transactions of the Chinese Society for Agricultural Machinery | 2019年 / 50卷 / 09期

关键词：

Digital elevation model; Elbow method; K-means clustering method; Linear regression; Radical basis function neural network; Unmanned aerial vehicle light detection and ranging;

D O I：

10.6041/j.issn.1000-1298.2019.09.024

中图分类号：

学科分类号：

摘要：

Digital elevation model (DEM) is a basic surface information product for constructing hydrological models, drawing slope maps, and extracting topographic features and so on. Because unmanned aerial vehicle (UAV) light detection and ranging (LiDAR) point cloud data has discrete characteristics, a reasonable interpolation method needs to be selected when generating DEM based on point clouds. The desert vegetation area in Xinjiang was taken as the research background, the zero-mean normalization method was used to normalize the point clouds' echo intensity, the elbow method was used to determine the optimal number of clustering by K-means approach, and the K-means clustering method was used to cluster the point clouds' intensity values to obtain the test area's ground point clouds. After that, the Kriging interpolation method was used to interpolate the ground point clouds with the thinning rate of 20% and 80%, respectively. Furthermore, the point clouds' elevation value was used as a variable to establish the radical basis function neural network (RBFNN) prediction model, the accuracy of RBFNN prediction model was analyzed by linear regression method, and then the high-precision DEM was generated by Delaunay triangulation interpolation. The results showed that K-means clustering method was adopted to realize the clustering with the optimal number of clustering as 4, and 48 722 ground point clouds were obtained. The root mean squared error (RMSE) corresponding to the point cloud thinning rate of 20% was smaller, and RBFNN training time was 56 s when the point cloud thinning rate was 20%. The determination coefficient R2 of fit for predicting the point clouds' elevation value was 0.887, and RMSE was 0.168 m when elevations of ground point clouds was predicted based on RBFNN. This method not only showed that the point cloud filtering can be realized by K-means clustering filtering, but also showed that the RBF neural network was a better way for predicting point cloud elevation. This can provide reference for constructing high-precision DEM based on point cloud. © 2019, Chinese Society of Agricultural Machinery. All right reserved.

引用

页码：208 / 214

页数：6

共 23 条

[1]

Polat N., Uysal M., Toprak A.S., An investigation of DEM generation process based on LiDAR data filtering, decimation, and interpolation methods for an urban area, Measurement, 75, pp. 50-56, (2015)

[2]

Ma D., Li B., Vegetation filtering in gully region of loess plateau based on laser scanning point cloud's intensity attenuation model and its terrain construct, Transactions of the CSAE, 29, 15, pp. 162-171, (2013)

[3]

Mao X., Hou J., Bai X., Et al., Multiscale forest gap segmentation and object-oriented classification based on DOM and LiDAR, Transactions of the Chinese Society for Agricultural Machinery, 48, 9, pp. 152-159, (2017)

[4]

Eitel J.U.H., Bernhard H., Vierling L.A., Et al., Beyond 3-D: the new spectrum of LiDAR applications for earth and ecological sciences, Remote Sensing of Environment, 186, pp. 372-392, (2016)

[5]

Lin X., Zhang J., Ning X., Et al., Filtering of point clouds using fusion of three types of primitives including points, objects and key points, Acta Geodaetica et Cartographica Sinica, 45, 11, pp. 1308-1317, (2016)

[6]

Li P., Xu Q., Xing S., Et al., Weighted curve fitting filtering method based on full-waveform LiDAR data, Geomatics and Information Science of Wuhan University, 43, 3, pp. 420-427, (2018)

[7]

Zhang J., Lin X., 3D power line reconstruction from airborne LiDAR point cloud of overhead electric power transmission corridors, Acta Geodaetica et Cartographica Sinica, 45, 3, pp. 347-353, (2016)

[8]

Zhou X., Research and application of airborne LiDAR point cloud data filters, (2011)

[9]

Wang Q., Researeh on the grid DEM interpolation algorithm of LiDAR points cloud, (2010)

[10]

Kucak R.A., Ozdemir E., Erol S., The segmentation of point clouds with K-means and ANN (artifical neural network), The International Archives of the Photogrammetry Remote Sensing and Spatial Information Sciences, pp. 595-598, (2017)

← 1 2 3 →