Feature Selection Algorithms of Airborne LiDAR Combined with Hyperspectral Images Based on XGBoost

Aiwu Zhang; Zhe Dong; Xiaoyan Kang

doi:10.3788/CJL201946.0404003

Chinese Journal of Lasers, Volume. 46, Issue 4, 0404003(2019)

Feature Selection Algorithms of Airborne LiDAR Combined with Hyperspectral Images Based on XGBoost

Aiwu Zhang^1,2、*, Zhe Dong^1,2, and Xiaoyan Kang^1,2

¹ Key Laboratory of 3D Information Acquisition and Application, Ministry of Education, Capital Normal University,Beijing 100048, China

² Engineering Research Center of Space Information Technology, Ministry of Education, Capital Normal University, Beijing 100048, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(10)

Fig. 1. Flow chart of XGB-SBS algorithm

Download full size

Fig. 2. Flow chart of XGB-PCCS algorithm

Download full size

Fig. 3. Map of research area. (a) LiDAR derived digital surface model; (b) hyperspectral image; (c) surface truth classification

Download full size

Fig. 4. Relationship between overall classification accuracy and number of feature set dimension retained by different selection algorithms

Download full size

Fig. 5. Absolute values of Pearson correlation coefficients between FGLDV_Con and other 45 kinds of features

Download full size

Table 1. Class name and number of samples in training and test sets

View table

Table 1. Class name and number of samples in training and test sets

No.	Class name	Number ofsample pointsin training set	Number ofsample pointsin test set
1	Healthy grass	2940	6859
2	Stressed grass	9751	22751
3	Artificial turf	205	479
4	Evergreen trees	4078	9517
5	Deciduous trees	1506	3515
6	Bare earth	1355	3161
7	Water	80	186
8	Residential buildings	11932	27840
9	Non-residential buildings	67125	156627
10	Roads	13760	32106
11	Sidewalks	10209	23820
12	Crosswalks	455	1063
13	Major thoroughfares	13904	32444
14	Highways	2959	6906
15	Railways	2081	4856
16	Paved parking lots	3450	8050
17	Unpaved parking lots	44	102
18	Cars	1964	4583
19	Trains	1611	3758
20	Stadium seats	2047	4777
	Total	151456	353400

Table 2. Optimal features at different feature selection algorithms

View table

Table 2. Optimal features at different feature selection algorithms

Algorithm	Number	Optimal features
XGB-SBS	33	F_H, F_I, F_NR, F_Z, F_Hm, F_Hvar, $\begin{matrix} {F_{S}}_{xy} \end{matrix}$ ,F_S_yz, F_VAm, F_Ns, $\begin{matrix} {F_{λ}}_{1} \end{matrix}$ , $\begin{matrix} {F_{λ}}_{2} \end{matrix}$ , $\begin{matrix} {F_{λ}}_{3} \end{matrix}$ , F_Ac, F_Pc, F_Sc, F_Ec,F_Oc,F_{PCA 1-10}, F_{PCA_m}, F_{PCA_var},F_{GLCM_Con}, F_{GLCM_Dis}, F_{GLCM_Cor}
XGB-PCCS	25	F_H, F_I, F_Z, F_Hm, F_Hvar, F_VA, F_VAm,F_Ns, F_Ac, F_Pc, F_Ec,F_{PCA 1-10}, F_{PCA_m}, F_{PCA_var},F_{GLCM_Con}, F_{GLCM_Cor}

Table 3. Order of feature deletion for finding optimal feature subset based on XGB-SBS algorithm
View table
Table 3. Order of feature deletion for finding optimal feature subset based on XGB-SBS algorithm
Order Deleted features
1-5 F_{GLDV_Con}, F_{GLDV_Mean}, F_{GLDV_ASM}, F_{GLCM_ASM}, F_{GLCM_Ent}
6-10 F_RN, $\begin{matrix} {F_{S}}_{xz} \end{matrix}$ , F_Lc, F_D, F_{GLCM_Hom}
11-13 F_VA, F_S, F_Ds

Table 4. Order of feature deletion for finding optimal feature subset based on XGB-PCCS algorithm

View table

Table 4. Order of feature deletion for finding optimal feature subset based on XGB-PCCS algorithm

Order	Deleted features
Order	According to feature importance	According to Pearson correlation coefficient
1	F_{GLDV_Con}	F_{GLCM_Hom}, F_{GLCM_Dis}, F_{GLCM_Ent}, F_{GLCM_ASM}, F_{GLDV_Mean}, F_{GLDV_ASM}
2	$\begin{matrix} {F_{S}}_{xz} \end{matrix}$	$\begin{matrix} {F_{λ}}_{1} \end{matrix}$ , $\begin{matrix} {F_{λ}}_{2} \end{matrix}$ , $\begin{matrix} {F_{λ}}_{3} \end{matrix}$ , $\begin{matrix} {F_{S}}_{yz} \end{matrix}$ , $\begin{matrix} {F_{S}}_{xy} \end{matrix}$ , F_RN, F_Oc, F_NR
3	F_Ds	F_S, F_Sc, F_Lc, F_D

Table 5. Running time consumption and predicted classification accuracy of two feature selection algorithms
View table
Table 5. Running time consumption and predicted classification accuracy of two feature selection algorithms
Result XGBoost XGB-SBS XGB-PCCS
Time consumption /s 17818 5255
Dimension offeature subset 46 33 25
OA /% 95.53 95.63 95.55
Kappa 0.942 0.943 0.942

Tools

Get Citation

Copy Citation Text

Aiwu Zhang, Zhe Dong, Xiaoyan Kang. Feature Selection Algorithms of Airborne LiDAR Combined with Hyperspectral Images Based on XGBoost[J]. Chinese Journal of Lasers, 2019, 46(4): 0404003

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: measurement and metrology

Received: Oct. 23, 2018

Accepted: Dec. 29, 2018

Published Online: May. 9, 2019

The Author Email:

DOI:10.3788/CJL201946.0404003

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Class name and number of samples in training and test sets

Table 1. Class name and number of samples in training and test sets

Table 2. Optimal features at different feature selection algorithms

Table 2. Optimal features at different feature selection algorithms

Table 3. Order of feature deletion for finding optimal feature subset based on XGB-SBS algorithm

Table 3. Order of feature deletion for finding optimal feature subset based on XGB-SBS algorithm

Table 4. Order of feature deletion for finding optimal feature subset based on XGB-PCCS algorithm

Table 4. Order of feature deletion for finding optimal feature subset based on XGB-PCCS algorithm

Table 5. Running time consumption and predicted classification accuracy of two feature selection algorithms

Table 5. Running time consumption and predicted classification accuracy of two feature selection algorithms