Hyperspectral Image Classification Based on Clustering Dimensionality Reduction and Visual Attention Mechanism

Chaoping Zeng; Lijun Ju; Jianchen Zhang

doi:10.3788/LOP56.212802

Journals >Laser & Optoelectronics Progress >Volume 56 >Issue 21 >Page 212802 > Article

Laser & Optoelectronics Progress
Vol. 56, Issue 21, 212802 (2019)

Hyperspectral Image Classification Based on Clustering Dimensionality Reduction and Visual Attention Mechanism

Chaoping Zeng¹, Lijun Ju¹, and Jianchen Zhang^2、*

Author Affiliations

¹Department of Space Information Engineering, Henan College of Surveying and Mapping, Zhengzhou, Henan 450015, China

²College of Environment and Planning, Henan University, Kaifeng, Henan 475004, China

show less

DOI: 10.3788/LOP56.212802 Cite this Article Set citation alerts

Chaoping Zeng, Lijun Ju, Jianchen Zhang. Hyperspectral Image Classification Based on Clustering Dimensionality Reduction and Visual Attention Mechanism[J]. Laser & Optoelectronics Progress, 2019, 56(21): 212802 Copy Citation Text

show less

Fig. 1. Indian hyperspectral image dataset. (a) False-color composite imge; (b) labeled data; (c) legends of classification

Download full size

Fig. 2. Pavia hyperspectral image dataset. (a) False-color composite imge; (b) labeled data; (c) legends of classificatin

Download full size

Fig. 3. Band mutual information matrix of hyperspectral datasets. (a) Indian dataset; (b) Pavia dataset

Download full size

Fig. 4. Results of saliency mapping of single band. (a) Band 122 and its saliency mapping selected by Indian dataset; (b) band 1 and its saliency mapping removed by Indian dataset; (c) band 94 and its saliency mapping selected by Pavia dataset; (d) band 1 and its saliency mapping removed by Pavia dataset

Download full size

Class	Class name	SVM	PCA	EAP	HSD5-SVM	HSD10-SVM
1	Alfalfa	0.8261	0.9706	0.9565	1.0000	0.9130
2	Corn-notill	0.6610	0.6808	0.7518	0.8504	0.8578
3	Corn-mintill	0.5584	0.6650	0.8563	0.9416	0.8718
4	Corn	0.3333	0.7200	0.9179	0.7156	0.8711
5	Grass-pasture	0.8170	0.8344	0.9316	0.9041	0.8388
6	Grass-trees	0.9408	0.7047	0.9614	0.8427	0.8990
7	Grass-pasture-mowed	0.8571	0.8750	0.9286	0.7857	0.9286
8	Hay-windrowed	0.8348	0.9914	0.9978	0.9604	1.0000
9	Oats	1.0000	1.0000	1.0000	1.0000	1.0000
10	Soybean-notill	0.6999	0.7917	0.7749	0.8657	0.8342
11	Soybean-mintill	0.7916	0.6025	0.7328	0.9357	0.9400
12	Soybean-clean	0.4742	0.7780	0.9343	0.8881	0.7336
13	Wheat	0.8872	0.9845	0.9771	0.9128	0.9436
14	Woods	0.9334	0.8771	0.9692	0.9992	0.9992
15	Buildings-grass-trees-drives	0.3569	0.8930	0.7640	0.9292	0.9373
16	Stone-steel-towers	0.9348	0.9753	0.9683	0.9565	1.0000
Kappa		0.6944	0.7126	0.8189	0.8970	0.8878
OA		0.7343	0.7430	0.8405	0.9099	0.9018
AA		0.7442	0.8340	0.9014	0.9055	0.9105
Time/s		16.70	3.29	4.21	5.70	6.99

Table 1. Evaluation indices of classification accuracy of different methods on Indian dataset (best results are highlighted in bold)

Class	Class name	SVM	PCA	EAP	HSD5-SVM	HSD10-SVM
1	Asphalt	0.9319	0.9541	0.9718	0.9783	0.9850
2	Meadows	0.9795	0.9583	0.9810	0.9938	0.9950
3	Gravel	0.7708	0.6850	0.9579	0.9786	0.9852
4	Trees	0.9334	0.8324	0.9043	0.8627	0.8597
5	Metal sheets	0.9915	0.9948	0.9875	0.9958	0.9815
6	Bare soil	0.8787	0.7153	0.9490	0.9919	0.9922
7	Bitumen	0.8492	0.5803	0.9955	0.9990	0.9888
8	Bricks	0.8892	0.7816	0.9842	0.9848	0.9932
9	Shadows	0.9987	0.9989	0.9991	0.9783	0.9560
Kappa		0.9144	0.8260	0.9609	0.9736	0.9752
OA		0.9357	0.8680	0.9705	0.9801	0.9813
AA		0.9136	0.8337	0.9700	0.9737	0.9707
Time/s		84.32	16.23	18.24	9.77	12.25

Table 2. Evaluation indices of classification accuracy of different methods on Pavia dataset (best results are highlighted in bold)

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information