Effective Feature Extraction and Identification Method Based on Tree Laser Point Cloud

Xiaoyi Lu; Ting Yun; Lianfeng Xue; Qiangfa Xu; Lin Cao

doi:10.3788/CJL201946.0510002

Chinese Journal of Lasers, Volume. 46, Issue 5, 0510002(2019)

Effective Feature Extraction and Identification Method Based on Tree Laser Point Cloud

Xiaoyi Lu¹, Ting Yun^1,2, Lianfeng Xue^1、*, Qiangfa Xu¹, and Lin Cao²

Author Affiliations

¹College of Information Science and Technology, Nanjing Forestory University, Nanjing, Jiangsu 210037, China

²Co-Innovation Center for Sustainable Forestry in Southern China, Nanjing Forestry University, Nanjing, Jiangsu 210037, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(18)

Fig. 1. Delineation effect of individual tree in an example area of Nanjing Forestry University

Download full size

Fig. 2. Scanned point clouds of sample trees. (a) Metasequoia; (b) palm; (c) spaindus; (d) rubber tree; (e) bamboo

Download full size

Fig. 3. Overview of work flow of tree species identification

Download full size

Fig. 4. Schematic of definition of partial parameters associated with apparent characteristics of trees. (a) Side view of sample tree (rubber tree); (b) top view of sample tree (rubber tree)

Download full size

Fig. 5. Schematic of acquisition of features related to grids in apparent characteristics of trees. (a) Top view of sample tree (rubber tree) divided into 8 segements and central vertical section of each segement; (b) schematic for meshing central vertical section

Download full size

Fig. 6. Sample trees (rubber tree and metasequoia) described based on results of V-feature. (a) Scanned point clouds of rubber tree and point cloud projections in the second and eighth rectangular layers; (b) scanned point clouds of metasequoia and point cloud projections in the second and eighth rectangular layers; (c) line chart indicating V-feature of each layer for rubber tree and metasequoia

Download full size

Fig. 7. Sample trees (rubber tree and metasequoia) described based on results of L-feature. (a) Point cloud projections and coordinate system representation regarding to L_k(r) function for rubber tree and metasequoia in the second rectangular layer; (b) point cloud projections and coordinate system representation regarding to L_k(r) function for rubber tree and metasequoia in the eighth rectangular

Download full size

Fig. 8. Boxplots of classification accuracies based on relative clustering characteristics. (a) Based on V-feature; (b) based on L_max-feature; (c) based on L_min-feature; (d) based on r_max-feature; (e) based on r_min-feature

Download full size

Fig. 9. Boxplots of classification accuracies based on features of point cloud distribution

Download full size

Fig. 10. Boxplots of classification accuracies based on apparent features

Download full size

Fig. 11. Boxplots of classification accuracies based on optimal features

Download full size

Fig. 12. Classification map of individual tree species in an example area of Nanjing Forestry University based on optimal features

Download full size

Table 1. Characterization of basic structures of sample trees in statistics

View table

Table 1. Characterization of basic structures of sample trees in statistics

Attribute	Metasequoia	Palm	Sapindus	Rubber tree	Bamboo
Tree height /m	28.8±4.1	18.4±1.8	16.2±2.3	15.6±2.9	13.7±3.6
Crown breadth /m	6.3±1.8	5.1±1.1	4.2±2.3	5.2±0.5	2.81±0.7
Crown volume /m³	463.8±48.4	163.9±23.4	243.2±28.6	186.3±18.2	106.7±13.8
Number of points cloud	187556±9773	105539±4184	142317±6251	120843±5735	88496±3857
Number of trees	48	54	42	51	45

Table 2. Parameters associated with relative clustering characteristics of trees
View table
Table 2. Parameters associated with relative clustering characteristics of trees
Feature name Denotation Number of features
V V₁,…,V₁₀ 10
L_max L_{max_1},…,L_{max_10} 10
L_min L_{min_1},…,L_{min_10} 10
r_max r_{max_1},…,r_{max_10} 10
r_min r_{min_1},…,r_{min_10} 10

Table 3. Classification results of cross-validation based on relative clustering features of {V₂,V₄,V₈,L
View table
Table 3. Classification results of cross-validation based on relative clustering features of {V₂,V₄,V₈,L
Dataset Metasequoia Palm Sapindus Rubbertree Bamboo Total Misjudgementrate /% Leakagerate /%
Metasequoia 8 2 2 3 2 17 52.9 50
Palm 3 8 2 3 3 19 57.9 55.5
Sapindus 2 3 6 3 1 15 60 57.2
Rubber tree 2 3 3 7 1 16 56.3 58.8
Bamboo 1 2 1 1 7 12 41.7 53.3
Correct number 8 8 6 7 7 36
Number of training samples 16 18 14 17 15 80
Accuracy /% 50 44.5 42.8 41.2 46.7 45

Table 4. Classification results of cross-validation based on features of point cloud distribution of {P_H,R_H40%-60%,R_H60%-80%}

View table

Table 4. Classification results of cross-validation based on features of point cloud distribution of {P_H,R_H40%-60%,R_H60%-80%}

Dataset	Metasequoia	Palm	Sapindus	Rubbertree	Bamboo	Total	Misjudgementrate /%	Leakagerate /%
Metasequoia	10	1	1	1	0	13	23.1	37.5
Palm	1	11	1	2	1	16	31.3	38.9
Sapindus	2	2	8	4	2	18	55.6	42.9
Rubber tree	2	2	3	9	3	19	52.6	47.1
Bamboo	1	2	1	1	9	14	35.7	40
Correct number	10	11	8	9	9	47
Number of training samples	16	18	14	17	15	80
Accuracy /%	62.5	61.1	57.1	52.9	60	58.8

Table 5. Classification results of cross-validation based on apparent features of {R_C/D,η_crown,Q_L}

View table

Table 5. Classification results of cross-validation based on apparent features of {R_C/D,η_crown,Q_L}

Dataset	Metasequoia	Palm	Sapindus	Rubbertree	Bamboo	Total	Misjudgementrate /%	Leakagerate /%
Metasequoia	12	0	1	1	1	15	20	25
Palm	1	11	3	1	1	17	35.3	38.9
Sapindus	1	2	9	2	1	15	40	42.9
Rubber tree	2	2	1	10	3	18	44.5	47.1
Bamboo	0	3	0	3	9	15	40	40
Correct number	12	11	9	10	9	51
Number of training samples	16	18	14	17	15	80
Accuracy /%	75	61.1	64.3	58.8	60	63.8

Table 6. Classification results of cross-validation based on optimal features of {V₂,V₄,V₈,L_{max_2}

View table

Table 6. Classification results of cross-validation based on optimal features of {V₂,V₄,V₈,L_{max_2}

Dataset	Metasequoia	Palm	Sapindus	Rubbertree	Bamboo	Total	Misjudgementrate /%	Leakagerate /%
Metasequoia	29	2	0	0	0	31	6.5	9.4
Palm	2	31	2	1	0	36	13.9	13.9
Sapindus	0	2	24	1	1	28	14.3	14.3
Rubber tree	1	1	2	29	2	35	17.1	14.7
Bamboo	0	0	0	3	27	30	10	10
Correct number	29	31	24	29	27	140
Number of training samples	32	36	28	34	30	160
Accuracy /%	90.6	86.1	85.7	85.3	90	87.5

Tools

Get Citation

Copy Citation Text

Xiaoyi Lu, Ting Yun, Lianfeng Xue, Qiangfa Xu, Lin Cao. Effective Feature Extraction and Identification Method Based on Tree Laser Point Cloud[J]. Chinese Journal of Lasers, 2019, 46(5): 0510002

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: Dec. 4, 2018

Accepted: Feb. 18, 2019

Published Online: Nov. 11, 2019

The Author Email: Xue Lianfeng (xuelianfeng@njfu.edu.cn)

DOI:10.3788/CJL201946.0510002

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Characterization of basic structures of sample trees in statistics

Table 1. Characterization of basic structures of sample trees in statistics

Table 2. Parameters associated with relative clustering characteristics of trees

Table 2. Parameters associated with relative clustering characteristics of trees

Table 4. Classification results of cross-validation based on features of point cloud distribution of {PH,RH40%-60%,RH60%-80%}

Table 4. Classification results of cross-validation based on features of point cloud distribution of {PH,RH40%-60%,RH60%-80%}

Table 5. Classification results of cross-validation based on apparent features of {RC/D,ηcrown,QL}

Table 5. Classification results of cross-validation based on apparent features of {RC/D,ηcrown,QL}

Table 6. Classification results of cross-validation based on optimal features of {V2,V4,V8,Lmax_2

Table 6. Classification results of cross-validation based on optimal features of {V2,V4,V8,Lmax_2

Table 4. Classification results of cross-validation based on features of point cloud distribution of {P_H,R_H40%-60%,R_H60%-80%}

Table 4. Classification results of cross-validation based on features of point cloud distribution of {P_H,R_H40%-60%,R_H60%-80%}

Table 5. Classification results of cross-validation based on apparent features of {R_C/D,η_crown,Q_L}

Table 5. Classification results of cross-validation based on apparent features of {R_C/D,η_crown,Q_L}

Table 6. Classification results of cross-validation based on optimal features of {V₂,V₄,V₈,L_{max_2}

Table 6. Classification results of cross-validation based on optimal features of {V₂,V₄,V₈,L_{max_2}