Spatially-Regularized Manifold Discriminant Analysis Algorithm for Hyperspectral Image Classification

Hong Huang; Lihua Wang; Guangyao Shi

doi:10.3788/AOS202040.0228001

Acta Optica Sinica, Volume. 40, Issue 2, 0228001(2020)

Spatially-Regularized Manifold Discriminant Analysis Algorithm for Hyperspectral Image Classification

Hong Huang^*, Lihua Wang, and Guangyao Shi

Key Laboratory of Optoelectronic Technique System of the Ministry of Education, Chongqing University, Chongqing 400044, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(12)

Fig. 1. Flow chart of SSRMDA algorithm

Download full size

Fig. 2. ERS segmentation and ground-truth images. (a) ERS segmentation image; (b) ground-truth image

Download full size

Fig. 3. Hyperspectral images in Indian Pines dataset. (a) False-color image; (b) ground-truth image

Download full size

Fig. 4. Hyperspectral images in Washington DC Mall dataset. (a) False-color image; (b) ground-truth image

Download full size

Fig. 5. Overall classification accuracy of SSRMDA algorithm with different values of K and K_b on different datasets. (a) Indian Pines dataset; (b) Washington DC Mall dataset

Download full size

Fig. 6. Overall classification accuracy of SSRMDA algorithm with different α values

Download full size

Fig. 7. Classification diagrams of different algorithms on Indian Pines dataset

Download full size

Fig. 8. Classification diagrams of different algorithms on Washington DC Mall dataset

Download full size

Table 1. Classification accuracy of different algorithms on Indian Pines dataset (number before ± indicates overall classification accuracy, and the unit is %; number after ± indicates standard deviation of ov

View table

Table 1. Classification accuracy of different algorithms on Indian Pines dataset (number before ± indicates overall classification accuracy, and the unit is %; number after ± indicates standard deviation of ov

Algorithm	n_i=5	n_i=10	n_i=15	n_i=20	n_i=30
RAW	51.81±2.37(0.463)	59.48±1.72(0.547)	65.43±1.39(0.613)	68.45±1.03(0.645)	71.72±1.04(0.681)
PCA	51.73±2.39(0.462)	59.30±1.71(0.545)	65.17±1.36(0.610)	68.22±0.98(0.643)	71.42±1.05(0.678)
NPE	50.17±2.32(0.445)	56.65±1.81(0.516)	61.86±1.67(0.574)	64.78±1.35(0.605)	67.02±1.24(0.629)
LPP	52.08±2.09(0.467)	59.47±1.80(0.547)	64.87±1.68(0.607)	67.13±1.35(0.631)	70.10±1.08(0.663)
LDA	54.82±2.38(0.498)	63.71±1.70(0.595)	69.87±1.55(0.662)	72.26±1.06(0.688)	75.12±1.10(0.719)
MFA	62.10±3.79(0.577)	74.82±2.03(0.716)	80.05±1.35(0.775)	82.99±1.20(0.807)	86.51±1.09(0.846)
LGSFA	60.80±4.10(0.563)	72.56±1.94(0.691)	80.38±1.62(0.778)	83.71±0.99(0.815)	88.05±1.20(0.864)
DSSM	54.15±3.30(0.490)	62.63±2.95(0.582)	72.08±1.51(0.686)	74.44±1.23(0.712)	76.91±1.17(0.739)
LPNPE	56.81±3.37(0.522)	70.14±2.27(0.666)	76.89±1.60(0.740)	81.97±1.55(0.796)	88.26±0.94(0.866)
SSRLDE	56.85±3.19(0.523)	71.38±2.80(0.680)	78.97±2.64(0.763)	83.34±1.97(0.812)	89.46±0.99(0.880)
SSRMDA	70.80±2.87(0.672)	82.14±2.65(0.798)	86.22±1.82(0.844)	89.21±1.51(0.876)	91.58±1.34(0.904)

Table 2. Classification accuracy of each-class grand object on Indian Pines dataset obtained by different algorithms%

View table

Table 2. Classification accuracy of each-class grand object on Indian Pines dataset obtained by different algorithms%

Class	RAW	PCA	NPE	LPP	LDA	MFA	LGSFA	DSSM	LPNPE	SSRLDE	SSRMDA
Alfalfa	97.22	97.22	97.22	97.22	100	100	100	97.22	100	100	100
Corn-Notill	61.33	61.11	55.82	60.25	69.26	83.56	82.63	59.11	76.55	79.63	86.92
Corn-mintill	58.55	58.55	52.89	54.73	67.52	75.89	76.99	56.58	76.26	77.61	89.54
Corn	53.30	53.74	43.61	48.01	67.84	92.07	93.83	44.05	85.90	90.31	88.99
Grass-pasture	71.46	71.24	63.63	67.44	83.51	89.21	89.43	73.57	89.21	89.85	91.12
Grass-trees	90.49	90.35	85.17	82.51	97.90	99.44	99.58	93.98	82.93	79.86	99.58
Grassp-asture-mowed	100	100	100	100	100	100	100	100	100	100	100
Hay-windrowed	87.82	87.82	77.35	81.19	71.58	97.43	97.65	88.03	92.52	98.50	97.43
Oats	100	100	100	100	100	100	100	100	100	100	100
Soybean-nottill	77.54	76.81	69.04	70.62	75.34	89.08	85.41	80.06	84.89	86.46	86.25
Soybean-mintill	74.93	74.73	68.70	70.15	73.73	83.41	85.82	72.19	81.21	83.41	92.02
Soybean-clean	47.33	46.64	43.71	46.12	46.47	86.23	78.31	43.37	71.77	83.821	90.88
Wheat	96.92	96.92	96.92	97.94	97.94	98.97	98.97	96.92	98.46	98.97	98.97
Woods	76.53	76.21	73.95	75.56	78.30	82.90	84.27	76.53	84.19	85.80	85.08
Buildings-Grass-Trees-Drives	71.01	71.27	68.88	69.15	76.59	85.63	84.84	72.87	91.49	88.03	86.97
Stone-Steel-Towers	100	100	97.59	100	100	100	100	98.79	100	100	100
OA	72.24	72.01	66.61	68.54	74.66	86.37	86.31	71.33	82.46	84.66	90.37
AA	79.03	78.91	74.65	76.31	81.62	91.49	91.11	78.33	88.46	90.14	93.36
Kappa	0.684	0.682	0.620	0.642	0.712	0.845	0.844	0.674	0.800	0.826	0.890

Table 3. Classification accuracy of different algorithms on Washington DC Mall dataset (number before ± indicates overall classification accuracy, and the unit is %; number after ± indicates standard deviation

View table

Table 3. Classification accuracy of different algorithms on Washington DC Mall dataset (number before ± indicates overall classification accuracy, and the unit is %; number after ± indicates standard deviation

Algorithm	n_i=5	n_i=10	n_i=15	n_i=20	n_i=30
RAW	80.86±2.96(0.761)	82.80±2.23(0.786)	84.07±2.17(0.802)	85.87±1.86(0.824)	86.60±1.22(0.833)
PCA	80.86±2.96(0.761)	82.79±2.23(0.786)	84.06±2.18(0.802)	85.86±1.86(0.824)	86.58±1.23(0.833)
NPE	79.91±3.33(0.750)	81.20±3.06(0.767)	82.91±2.13(0.788)	84.94±1.74(0.812)	85.28±1.67(0.817)
LPP	80.21±3.35(0.753)	82.24±1.91(0.779)	83.47±1.87(0.795)	85.46±1.81(0.819)	86.39±1.72(0.831)
LDA	81.56±3.33(0.770)	83.08±2.23(0.790)	85.52±1.77(0.820)	86.69±1.48(0.834)	87.84±1.27(0.848)
MFA	84.49±2.51(0.806)	87.92±2.25(0.849)	90.98±2.20(0.887)	91.87±0.89(0.898)	92.70±0.59(0.908)
LGSFA	85.48±2.68(0.818)	88.53±2.20(0.857)	91.28±1.15(0.891)	92.47±1.03(0.906)	93.64±0.98(0.920)
DSSM	80.63±3.31(0.759)	82.80±2.19(0.786)	84.08±2.57(0.802)	85.33±2.11(0.817)	86.61±1.22(0.833)
LPNPE	78.71±5.15(0.736)	86.65±2.63(0.834)	87.75±2.36(0.847)	89.14±1.02(0.864)	90.34±1.06(0.879)
SSRLDE	79.45±5.36(0.744)	86.51±4.22(0.832)	87.34±3.30(0.843)	89.62±1.37(0.870)	90.86±1.19(0.886)
SSRMDA	88.73±2.15(0.859)	92.78±1.19(0.910)	94.04±0.98(0.925)	95.35±0.73(0.942)	96.67±0.73(0.958)

Table 4. Classification accuracy of each-class ground object on Washington DC Mall dataset obtained by different algorithms%

View table

Table 4. Classification accuracy of each-class ground object on Washington DC Mall dataset obtained by different algorithms%

Class	RAW	PCA	NPE	LPP	LDA	MFA	LGSFA	DSSM	LPNPE	SSRLDE	SSRMDA
Road	95.04	95.01	94.71	94.38	94.76	97.83	98.71	95.17	93.72	95.06	99.79
Water	94.12	94.12	93.57	94.74	96.33	97.24	98.22	94.13	99.26	96.45	98.39
Building	87.67	87.64	86.56	86.53	88.33	91.40	93.15	87.64	96.93	95.39	98.61
Vegetation	97.34	97.34	97.01	97.27	97.37	97.88	97.88	97.34	97.08	96.40	98.73
Trail	66.83	66.89	64.04	69.39	72.35	89.52	93.20	66.66	91.86	82.38	93.76
Shadow	67.90	67.90	65.03	67.73	68.20	73.78	80.46	67.86	72.85	72.05	82.66
OA	88.09	88.08	87.05	87.91	88.87	92.81	94.64	88.11	92.90	91.64	96.70
AA	84.82	84.82	83.49	85.01	86.22	91.28	93.60	84.80	91.95	89.62	95.49
Kappa	0.850	0.849	0.837	0.848	0.860	0.909	0.933	0.850	0.911	0.895	0.959

Tools

Get Citation

Copy Citation Text

Hong Huang, Lihua Wang, Guangyao Shi. Spatially-Regularized Manifold Discriminant Analysis Algorithm for Hyperspectral Image Classification[J]. Acta Optica Sinica, 2020, 40(2): 0228001

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: Jul. 18, 2019

Accepted: Sep. 6, 2019

Published Online: Jan. 2, 2020

The Author Email: Huang Hong (hhuang@cqu.edu.cn)

DOI:10.3788/AOS202040.0228001

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology