Study on Classification of Arbor Tree Species at Single Tree Scale based on Cross-modal Hybrid Fusion of UAV Point Cloud and Image

指数特征变量	变量描述	参考文献	指数特征变量	变量描述	参考文献
红(R)	R=R	—	最大值	$V_{m a x} = M A X (X)$	—
绿(G)	G=G	—	最小值	$V_{m i n} = M I N (X)$	—
蓝(B)	B=B	—	平均数	$V m e a n = (\sum_{i = 1}^{n} x_{i}) / n$	—
绿叶指数(GLI)	$G L I = \frac{((G - R) + (G - B))}{(2 * G + R + B)}$	[20]	方差	$V_{v a r i a n c e} = \sqrt[]{\frac{\sum_{i = 1}^{n} {(x_{i} - M e a n)}^{2}}{n}}$	—
绿红植被指数(GRVI)	$G R V I = \frac{R - G}{R + G}$	[20]	偏斜度	$V_{s k e w n e s s} = \sqrt[3]{\frac{\sum_{i = 1}^{n} {(x_{i} - M e a n)}^{3}}{n}}$	—
红绿蓝植被指数(RGBVI)	$R G B V I = \frac{G * G - R * B}{G * G + R * B}$	[20]	冠幅面积(S)	$S = \sum_{i = 1}^{n} p_{i}$	—
可见大气阻力指数(VARI)	$V A R I = \frac{G - R}{G + R - B}$	[20]	归一化绿红差异指数(NGRDI)	$N G R D I = \frac{G - R}{G + R}$	[20]

Table 4. Point cloud parameter characteristic variable and its description

View table

View in Article

Table 4. Point cloud parameter characteristic variable and its description

二次幂平均(1个)

Q P M = \sqrt[]{\sum_{i = 1}^{n} Z_{i}^{2} / n}

强度变量(42个)

强度百分位数四分位数间距(1个)

I n t_I Q = I n t_{75 %} - I n t_{25 %}

三次幂平均(1个)

T P M = \sqrt[3]{\sum_{i = 1}^{n} Z_{i}^{3} / n}

郁闭度

(1个)

郁闭度(CC,1个)

$C C = n_{v e g} / n_{t o t a l}$

( n_veg为植被点数，n_total为总点数)

高度百分位数(15个)

$E l e v = Z_{X %}$ (X=1,5,10,20,

25,30,40,50,60,70,75,80,90,95,99)

叶面积

指数

(1个)

叶面积指数

(LAI, 1个)

L A I = (c o s (a n g) \times l n (G F)) / k

(ang为平均扫描角，GF为间隙率，k为消光系数)

高度百分位数四分位数间距(1个)

E l e v_I Q = E l e v_{75 %} - E l e v_{25 %}

间隙率

(1个)

间隙率(GF, 1个)

$G F = n_{g r o u n d} / n$ ( n_ground为提取Z值

低于高度阈值的地面点数，n为总点数)

Table 5. Parameter Settings of each algorithm

View table

View in Article

Table 5. Parameter Settings of each algorithm

SVM

XGBoost

LightGBM

c=15, 55(LMF)

g=0.01

max_depth =5

n_estimators=500

max_depth=5

learning_rate=0.1

n_estimators=500

max_depth=5

learning_rate=0.1

n_estimators=500

num_leaves=60

Bagging

AdaBoost

Blending

Stacking

base_estimator=LightGBM

n_estimators=500

max_samples=10

max_feature=10

base_estimator=LightGBM

n_estimators=500

learning_rate=0.1

base_estimator：

XGBoost、LightGBM、RF

meta_estimator：

Naive Bayes

base_estimator：

XGBoost、LightGBM、RF

meta _estimator：

Naive Bayes

Table 6. Performance comparison of each classification algorithm

View table

View in Article

Table 6. Performance comparison of each classification algorithm

名称	数据类型	测试精度/%	Kappa	用时/S
SVM	Point Cloud	78.04	0.722 2	0.012
RF	Point Cloud	68.86	0.604 3	0.016
SVM	Image	80.44	0.751 7	0.011
RF	Image	75.05	0.683 0	0.013
SVM	Concatenate	86.63	0.830 5	0.014
RF	Concatenate	76.85	0.707 0	0.039
SVM	LMF	91.15	0.888 1	0.005
RF	LMF	89.02	0.861 4	0.103
XGBoost	LMF	94.14	0.925 4	0.509
LightGBM	LMF	95.44	0.942 0	0.212
Proposed Method	Proposed	99.40	0.992 4	1.23

Table 7. Performance comparison of different ensemble methods
View table
View in Article
Table 7. Performance comparison of different ensemble methods
名称数据类型测试精度/% Kappa 用时/S
Bagging Concatenate 96.01 0.949 4 15.510
AdaBoost Concatenate 97.21 0.964 5 2.215
Blending Proposed 94.42 0.928 7 1.772
Stacking Proposed 99.40 0.992 4 1.23

Table 8. Comparison of tree species classification performance of CNN algorithm

View table

View in Article

Table 8. Comparison of tree species classification performance of CNN algorithm

名称	模型深度	验证精度 /%	测试精度 /%	测试集 Kappa	训练用时
Alex Net	8	88.96	88.13	0.848 6	52 s
Google Net	22	87.66	86.30	0.826 5	1 min 19 s
VGG-16 Net	16	82.47	81.74	0.769 5	10 min 2 s
Res Net-18	18	91.56	90.41	0.878 4	1 min 14 s
Dense Net	201	94.81	92.24	0.901 3	23 min 24 s
Shuffle Net	50	87.66	84.02	0.799 4	2 min 27 s

Table 9. Statistical results of tree in study area
View table
View in Article
Table 9. Statistical results of tree in study area
柏树杜仲油橄榄松树杂木总数
388 139 524 343 1 096 2 490

Tools

Get Citation

Copy Citation Text

Min YAN, Yonghua XIA, Chong WANG, Xiali KONG, Haoyu TAI, Chen LI. Study on Classification of Arbor Tree Species at Single Tree Scale based on Cross-modal Hybrid Fusion of UAV Point Cloud and Image[J]. Remote Sensing Technology and Application, 2024, 39(1): 87

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites