Autoencoder-Based Fusion Classification of Hyperspectral and LiDAR Data

Fig. 4. Visualization of the Houston dataset. (a) Pseudocolor image for the HIS; (b) gray-scale image for the LiDAR; (c) training data; (d) testing data

Download full size

Fig. 5. Visualization of the Trento dataset. (a) Pseudocolor image for the HIS; (b) gray-scale image for the LiDAR; (c) training data; (d) testing data

Download full size

Fig. 6. Houston data classification results. (a) Color composite image; (b) (1) (H+L); (c) (2) (H+L); (d) (3) (H+L); (e) (4) (H+L); (f) (5) (L); (g) (5) (H); (h) (5) (H+L)

Download full size

Fig. 7. Trento data classification results. (a) Color composite image; (b) (1) (H+L); (c) (2) (H+L); (d) (3) (H+L); (e) (4) (H+L); (f) (5) (L); (g) (5) (H); (h) (5) (H+L)

Download full size

Table 1. Description table of the parameters of the Houston and Trento dataset

View table

Table 1. Description table of the parameters of the Houston and Trento dataset

Dataset	Houston			Trento
Location	Houston，Texas，USA			Trento，Italy
Sensor type	HSI		LiDAR-based DSM	HSI	LiDAR-based DSM
Image size /pixel	349×1905			600×166
Spatial resolution /m	2.5			1
Number of bands	144	1		63	1
Sensor name	ITRES CASI 1500			AISA Eagle	Optech ALTM 3100EA

Table 2. Number of training and testing samples for different feature classes in the Houston dataset

View table

Table 2. Number of training and testing samples for different feature classes in the Houston dataset

No.	Class	Train	Test
Total		2832	15029
1	health grass	198	1251
2	stressed grass	190	1254
3	synthetic grass	192	697
4	tress	188	1244
5	soil	186	1242
6	water	182	325
7	residential	196	1268
8	commercial	191	1244
9	road	193	1252
10	highway	191	1227
11	railway	181	1235
12	parking lot 1	192	1233
13	parking lot 2	184	469
14	tennis court	181	428
15	running track	187	660

Table 3. Number of training and testing samples for different feature classes in the Trento dataset
View table
Table 3. Number of training and testing samples for different feature classes in the Trento dataset
No. Class Train Test
Total 819 30214
1 apple trees 129 4034
2 buildings 125 2903
3 ground 105 479
4 woods 154 9123
5 vineyard 184 10501
6 roads 122 3174

Table 4. Comparison table of results of ablation experiment evaluation
View table
Table 4. Comparison table of results of ablation experiment evaluation
Dataset CNN AE A_OA /％ A_AA /％ Kappa
Houston √ ‒ 87.98 90.11 0.8698
‒ √ 88.52 89.95 0.8759
√ √ 93.66 94.20 0.9316
Trento √ ‒ 97.92 96.19 0.9681
‒ √ 94.17 93.88 92.22
√ √ 98.58 97.74 0.9811

Table 5. Classification accuracy of different models on Houston data

View table

Table 5. Classification accuracy of different models on Houston data

No.	Class	（1）（H+L）	（2）（H+L）	（3）（H+L）	（4）（H+L）	（5）（L）	（5）（H）	（5）（H+L）
1	health grass	82.43	83.10	81.58	94.49	59.35	82.81	82.43
2	stressed grass	82.05	83.70	83.65	92.78	26.41	85.15	97.27
3	synthetic grass	99.80	100.00	100.00	94.90	38.02	96.44	97.43
4	tress	92.80	91.86	93.09	93.20	77.94	90.72	99.53
5	soil	98.48	98.86	99.91	94.80	31.91	100.00	100.00
6	water	95.10	95.10	95.10	86.08	37.76	100.00	97.20
7	residential	75.47	80.04	82.65	89.42	80.88	87.50	95.06
8	commercial	46.91	68.47	81.29	85.80	83.57	90.50	96.49
9	road	79.53	84.80	88.29	86.34	51.09	84.04	93.30
10	highway	60.04	49.13	89.00	86.94	53.19	67.08	73.36
11	railway	81.02	80.27	83.78	90.60	86.15	94.97	96.58
12	parking lot 1	85.49	79.06	90.39	85.31	51.30	96.15	98.08
13	parking lot 2	75.09	71.58	82.46	87.40	70.53	90.52	94.74
14	tennis court	100.00	99.60	100.00	84.49	74.49	99.60	95.14
15	running track	98.31	98.52	98.10	94.72	46.30	98.31	96.41
A_OA /％		80.49	81.92	88.52	90.14	59.01	89.11	93.66
A_AA /％		83.50	84.27	89.95	89.82	57.93	90.92	94.20
Kappa		0.7898	0.8045	0.8759	0.8973	0.5698	0.8829	0.9316

Table 6. Classification accuracy of different models on Trento data

View table

Table 6. Classification accuracy of different models on Trento data

No.	Class	（1）（H+L）	（2）（H+L）	（3）（H+L）	（4）（H+L）	（5）（L）	（5）（H）	（5）（H+L）
1	apple trees	88.62	95.81	88.19	96.01	99.69	99.62	99.61
2	buildings	94.04	96.97	98.49	88.64	82.83	93.52	97.91
3	ground	93.53	96.66	95.19	98.02	40.64	91.44	96.62
4	woods	98.90	99.39	99.30	99.68	97.77	99.61	99.26
5	vineyard	88.96	82.24	91.96	99.43	93.75	99.91	99.97
6	roads	91.75	86.52	90.14	93.51	59.14	82.73	93.09
A_OA /％		92.77	91.32	94.17	97.36	90.46	97.29	98.58
A_AA /％		92.63	92.93	93.88	95.88	78.97	94.47	97.74
Kappa		0.9585	0.9042	0.9222	0.9646	0.8754	0.9639	0.9811

Table 7. Performance comparison of the proposed method applied to different training sample sizes
View table
Table 7. Performance comparison of the proposed method applied to different training sample sizes
Dataset Training sample rate A_OA/％ A_AA/％ Kappa
Houston 0.1 91.93 90.24 0.9102
0.15 92.20 91.77 0.9155
0.2 94.25 93.80 0.9377
Trento 0.1 99.31 99.28 0.9908
0.15 99.48 99.39 0.9931
0.2 99.64 99.49 0.9952

Tools

Get Citation

Copy Citation Text

Yibo Wang, Song Dai, Dongmei Song, Guofa Cao, Jie Ren. Autoencoder-Based Fusion Classification of Hyperspectral and LiDAR Data[J]. Laser & Optoelectronics Progress, 2024, 61(12): 1228001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Remote Sensing and Sensors

Received: May. 9, 2023

Accepted: Aug. 10, 2023

Published Online: Jun. 17, 2024

The Author Email: Song Dai (b22160011@s.upc.edu.cn)

DOI:10.3788/LOP231262

CSTR:32186.14.LOP231262

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Description table of the parameters of the Houston and Trento dataset

Table 1. Description table of the parameters of the Houston and Trento dataset

Table 2. Number of training and testing samples for different feature classes in the Houston dataset

Table 2. Number of training and testing samples for different feature classes in the Houston dataset

Table 3. Number of training and testing samples for different feature classes in the Trento dataset

Table 3. Number of training and testing samples for different feature classes in the Trento dataset

Table 4. Comparison table of results of ablation experiment evaluation

Table 4. Comparison table of results of ablation experiment evaluation

Table 5. Classification accuracy of different models on Houston data

Table 5. Classification accuracy of different models on Houston data

Table 6. Classification accuracy of different models on Trento data

Table 6. Classification accuracy of different models on Trento data

Table 7. Performance comparison of the proposed method applied to different training sample sizes

Table 7. Performance comparison of the proposed method applied to different training sample sizes

微信扫一扫：分享