Airborne LiDAR Point Cloud Classification Based on Multiple-Entity Eigenvector Fusion

Hu Haiying; Hui Zhenyang; Li Na

doi:10.3788/CJL202047.0810002

Chinese Journal of Lasers, Volume. 47, Issue 8, 810002(2020)

Airborne LiDAR Point Cloud Classification Based on Multiple-Entity Eigenvector Fusion

Hu Haiying^1,2, Hui Zhenyang^1,2、*, and Li Na^1,2

Author Affiliations

¹Key laboratory of Digital Land and Resources, East China University of Technology, Nanchang, Jiangxi 330013, China

²Faculty of Geomatics, East China University of Technology, Nanchang, Jiangxi 330013, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Flow chart of point cloud classification

Download full size

Fig. 2. Flow chart of extraction of object entities

Download full size

Fig. 3. Diagram of extraction of object entities. (a) Raw data; (b) data after extraction of object entities

Download full size

Fig. 4. Rasterization of building point cloud. (a) Projection of building point cloud; (b) rasterization result when grid size is 0.5 m; (c) rasterization result when grid size is 1 m

Download full size

Fig. 5. Diagram of maximum bounding rectangles of building point cloud before and after rotation. (a) Building point cloud projection; (b) building point cloud projection after rotation

Download full size

Fig. 6. Diagram of experimental data. (a) Ankeny; (b) diagram of artificial classification of Ankeny; (c) Building; (d) diagram of artificial classification of Building; (e) Cadastre; (f) diagram of artificial classification of Cadastre

Download full size

Fig. 7. Classification errors using different eigenvector sets

Download full size

Table 1. Eigenvector size based on eigenvalues
View table
Table 1. Eigenvector size based on eigenvalues
Eigenvector Size
Linearity V₁=(λ₁-λ₂)/λ₁
Planarity V₂=(λ₂-λ₃)/λ₁
Scatter V₃=λ₃/λ₁
Anisotropy V₄=(λ₁-λ₃)/λ₁
Eigenentropy $\begin{matrix} V_{5} = - \overset{3}{\sum_{i' = 1}} λ_{i'} \times \ln (λ_{i'}) \end{matrix}$
Omnivariance V₆= $\begin{matrix} \sqrt[3]{λ_{1} \times λ_{2} \times λ_{3}} \end{matrix}$
Surface variation V₇=λ₃

Table 2. Eigenvector size based on elevation information

View table

Table 2. Eigenvector size based on elevation information

Eigenvector	Size
Height above	V₈=z_max(p_i)-z_p
Height below	V₉=z_p-z_min(p_i)
Height average	$\begin{matrix} V_{10} = \overset{k}{\sum_{i = 1}} z (p_{i} \frac{)}{k} \end{matrix}$
Vertical Range	V₁₁=z_max(p_i)-z_min(p_i)
Height standarddeviation	$\begin{matrix} V_{12} = \sqrt[]{\frac{1}{k} \overset{k}{\sum_{i = 1}} [z(p_{i}) - V_{10}]^{2}} \end{matrix}$
Height kurtosis	$\begin{matrix} V_{13} = \frac{\overset{k}{\sum_{i = 1}} [z(p_{i}) - V_{10}]^{3}}{{\{\overset{k}{\sum_{i = 1}} [z(p_{i}) - V_{10}]^{2}\}}^{\frac{3}{2}}} \end{matrix}$
Height skewness	$\begin{matrix} V_{14} = \frac{\overset{k}{\sum_{i = 1}} [z(p_{i}) - V_{10}]^{4}}{{\{\overset{k}{\sum_{i = 1}} [z(p_{i}) - V_{10}]^{2}\}}^{2} - 3} \end{matrix}$

Table 3. Eigenvector size based on surface information
View table
Table 3. Eigenvector size based on surface information
Eigenvector Size
Plane roughness V₁₅= $\begin{matrix} \frac{| A_{0} x_{0} + B_{0} y_{0} + C_{0} z_{0} + D_{0} |}{\sqrt[]{A_{0}^{2} + B_{0}^{2} + C_{0}^{2}}} \end{matrix}$
Plane range V₁₆=max(D'_i)
Plane standard deviation $\begin{matrix} V_{17} = \sqrt[]{\frac{1}{k} \overset{k}{\sum_{i = 1}} (D'_{i} {- \overset{̅}{D}')}^{2}} \end{matrix}$

Table 4. Evaluation index values of three classification methods based on Ankeny

View table

Table 4. Evaluation index values of three classification methods based on Ankeny

Category	Recall /%			Precision /%			F1 score /%
Category	RF	SVM	BP	RF	SVM	BP	RF	SVM	BP
Ground	96.28	96.44	11.93	85.07	84.65	47.13	0.90	0.90	0.19
High vegetation	44.70	43.34	95.08	83.57	78.74	78.62	0.58	0.56	0.86
Building	97.38	97.51	0	81.42	80.00	0	0.89	0.88	0
Road	67.80	64.40	0.01	95.49	96.72	1.83	0.79	0.77	0
Car	60.76	51.77	99.24	50.87	58.30	49.13	0.55	0.55	0.66
Human-made object	21.88	16.06	1.09	36.68	39.76	5.99	0.27	0.23	0.02
Mean	64.80	61.59	34.56	72.18	73.03	30.45	0.66	0.65	0.29

Table 5. Evaluation index values of three classification methods based on Building

View table

Table 5. Evaluation index values of three classification methods based on Building

Category	Recall /%			Precision /%			F1 score /%
Category	RF	SVM	BP	RF	SVM	BP	RF	SVM	BP
Ground	72.71	73.60	90.72	88.80	87.16	76.86	0.80	0.80	0.83
High vegetation	87.85	76.49	37.73	71.59	73.71	60.56	0.79	0.75	0.46
Building	91.88	97.17	2.03	94.88	84.52	18.72	0.93	0.90	0.04
Road	97.97	97.98	69.74	89.42	90.33	18.17	0.93	0.94	0.29
Car	50.92	7.18	30.04	54.69	60.45	78.75	0.53	0.13	0.43
Human-made object	5.43	0	0	13.00	0	0	0.08	0	0
Mean	67.79	58.74	38.37	68.73	66.03	42.18	0.68	0.59	0.34

Table 6. Evaluation index values of three classification methods based on Cadastre

View table

Table 6. Evaluation index values of three classification methods based on Cadastre

Category	Recall /%			Precision /%			F1 score /%
Category	RF	SVM	BP	RF	SVM	BP	RF	SVM	BP
Ground	75.87	74.76	2.32	89.69	91.50	7.94	0.82	0.82	0.04
High vegetation	82.46	80.94	65.68	67.34	70.16	59.14	0.74	0.75	0.62
Building	86.87	84.27	0	55.72	48.23	0	0.68	0.61	0
Road	77.53	79.75	10.34	90.80	87.15	15.30	0.84	0.83	0.12
Car	7.72	0	63.26	55.90	0.00	29.30	0.14	0	0.40
Human-made object	8.52	0.67	4.72	10.54	100.00	9.14	0.09	0.01	0.06
Mean	56.49	53.40	24.39	61.67	66.17	20.14	0.55	0.51	0.21

Tools

Get Citation

Copy Citation Text

Hu Haiying, Hui Zhenyang, Li Na. Airborne LiDAR Point Cloud Classification Based on Multiple-Entity Eigenvector Fusion[J]. Chinese Journal of Lasers, 2020, 47(8): 810002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: remote sensing and sensor

Received: Nov. 20, 2019

Accepted: --

Published Online: Aug. 17, 2020

The Author Email: Zhenyang Hui (huizhenyang2008@163.com)

DOI:10.3788/CJL202047.0810002

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Eigenvector size based on eigenvalues

Table 1. Eigenvector size based on eigenvalues

Table 2. Eigenvector size based on elevation information

Table 2. Eigenvector size based on elevation information

Table 3. Eigenvector size based on surface information

Table 3. Eigenvector size based on surface information

Table 4. Evaluation index values of three classification methods based on Ankeny

Table 4. Evaluation index values of three classification methods based on Ankeny

Table 5. Evaluation index values of three classification methods based on Building

Table 5. Evaluation index values of three classification methods based on Building

Table 6. Evaluation index values of three classification methods based on Cadastre

Table 6. Evaluation index values of three classification methods based on Cadastre