Feature Extraction of Hyperspectral Image Based on Sparse Representation and Learning Graph Regularity

Minghua ZHANG; Hongling LUO; Wei SONG; Dongmei HUANG; Qi HE; Cheng SU

doi:10.3788/gzxb20215004.0410002

Journals >Acta Photonica Sinica >Volume 50 >Issue 4 >Page 241 > Article

Acta Photonica Sinica
Vol. 50, Issue 4, 241 (2021)

Feature Extraction of Hyperspectral Image Based on Sparse Representation and Learning Graph Regularity

Minghua ZHANG¹, Hongling LUO¹, Wei SONG^1、*, Dongmei HUANG^1、2, Qi HE¹, and Cheng SU³

Author Affiliations

¹College of Information Technology， Shanghai Ocean University， Shanghai20306， China

²Shanghai University of Electric Power， Shanghai00090， China

³East China Sea Forecast Center， Ministry of Natural Resources， Shanghai20016， China

show less

DOI: 10.3788/gzxb20215004.0410002 Cite this Article

Minghua ZHANG, Hongling LUO, Wei SONG, Dongmei HUANG, Qi HE, Cheng SU. Feature Extraction of Hyperspectral Image Based on Sparse Representation and Learning Graph Regularity[J]. Acta Photonica Sinica, 2021, 50(4): 241 Copy Citation Text

show less

Fig. 1. Flowchart of algorithm LDGSPP

Download full size | View in the Article

Fig. 2. PaviaU dataset

Download full size | View in the Article

Fig. 3. Indian Pines dataset

Download full size | View in the Article

Fig. 4. Classification results of PaviaU dataset with different algorithms

Download full size | View in the Article

Fig. 5. 3D projection graph of PaviaU dataset with different algorithms

Download full size | View in the Article

Fig. 6. Classification results of Indian Pines dataset with different algorithms

Download full size | View in the Article

Fig. 7. Sensitivity analysis of regular parameters ρ

Download full size | View in the Article

Fig. 8. Sensitivity analysis of subspace dimension

Download full size | View in the Article

Fig. 9. Sensitivity analysis of training set size

Download full size | View in the Article

Fig. 10. Convergence curve

Download full size | View in the Article

	1⁃NN									SVM
Method	OSF	PCA	LPP	LDMGI	SPA	MSME	LESR	LMSDL	LDGSPP	OSF	PCA	LPP	LDMGI	SPA	MSME	LESR	LMSDL	LDGSPP
Class1	70.2	70.4	74.2	73.8	71.0	69.2	74.4	81.8	78.2	75.8	78.2	81.5	78.9	77.8	74.6	79.3	89.7	92.3
Class2	70.1	70.2	72.0	70.9	70.8	67.6	71.8	78.1	91.6	79.0	79.5	70.5	69.6	80.9	70.4	85.2	82.1	94.9
Class3	68.3	68.2	73.0	65.0	70.1	63.8	67.5	78.1	80.9	84.4	82.5	80.9	73.7	83.7	82.1	84.7	84.6	78.4
Class4	90.9	90.9	90.6	93.4	91.5	91.0	92.3	92.5	93.1	92.4	90.1	90.8	96.0	92.6	93.2	94.2	96.2	96.3
Class5	98.9	98.9	99.8	99.9	98.9	98.4	99.5	100	100	99.5	99.5	100	99.9	99.6	99.5	99.6	100	100
Class6	70.0	69.9	75.1	70.3	67.4	67.7	71.1	78.7	94.8	80.7	82.9	83.2	89.0	83.7	79.8	85.5	84.6	97.7
Class7	90.5	90.2	83.2	89.8	89.4	87.2	88.7	92.5	97.8	92.2	92.0	81.1	91.8	93.1	91.7	92.9	91.8	98.5
Class8	70.5	70.5	69.0	72.1	69.2	67.5	70.4	77.0	77.1	76.7	76.6	75.8	77.9	75.8	75.8	80.7	83.2	88.6
Class9	99.8	99.9	99.9	99.8	99.8	100	99.8	99.7	100	99.5	99.5	99.7	99.8	99.7	99.9	99.7	99.8	100
AA（%）	81.0	81.0	81.9	81.7	80.8	79.2	81.7	86.5	90.4	86.7	86.8	84.8	86.3	87.4	85.2	89.1	90.2	94.1
OA（%）	73.6	73.7	75.6	74.7	73.7	71.6	75.2	81.3	88.8	81.2	81.8	77.9	78.3	82.6	76.9	85.5	86.0	94.0
KA（%）	66.5	66.6	68.9	60.8	66.6	64.0	68.4	75.9	85.4	75.8	76.5	72.1	72.7	77.6	70.8	81.2	81.9	92.0

Table 1. Classification performance of PaviaU dataset with various methods（%）

View in the Article

	1⁃NN									SVM
Method	OSF	PCA	LPP	LDMGI	SPA	MSME	LESR	LMSDL	LDGSPP	OSF	PCA	LPP	LDMGI	SPA	MSME	LESR	LMSDL	LDGSPP
Class1	48.3	41.9	48.4	41.9	64.5	48.4	64.5	64.5	77.4	35.4	48.4	58.1	64.5	80.6	61.3	80.6	64.5	87.1
Class2	43.1	41.8	43.9	35.2	43.8	41.5	43.5	59.4	60.8	52.1	51.6	60.3	47.2	61.5	54.9	60.5	75.4	85.2
Class3	53.3	52.9	53.4	51.8	51.4	52.9	55.9	62.7	66.6	59.8	58.7	56.6	59.3	60.0	53.1	68.9	78.1	83.2
Class4	66.6	70.1	65.5	61.0	66.1	63.3	62.1	76.2	92.6	77.9	75.1	73.4	74.5	75.1	74.0	75.1	87.0	96.0
Class5	90.3	90.3	89.8	84.1	91.3	88.9	81.0	93.8	95.9	90.3	91.5	94.3	90.5	92.4	92.0	87.9	95.2	95.9
Class6	91.6	89.0	92.4	85.5	88.1	90.0	89.7	92.5	98.5	91.9	91.2	90.0	88.6	91.9	90.7	92.8	97.4	99.4
Class7	92.3	84.6	92.3	84.6	92.3	92.3	61.5	84.6	100	92.3	92.3	92.3	92.3	92.3	92.3	84.6	84.6	100
Class8	88.7	87.3	84.9	87.0	86.6	84.9	92.1	93.5	94.7	95.2	94.7	88.0	89.2	95.0	95.5	96.6	98.3	99.7
Class9	80	80.0	80.0	100	80.0	60.0	100	100	100	100	100	100	100	100	100	100	100	100
Class10	64.4	63.2	60.0	58.9	62.7	62.6	65.4	76.6	70.1	67.5	68.9	61.4	72.0	65.1	63.7	75.5	80.0	83.0
Class11	47.2	46.9	46.7	46.6	47.4	46.2	47.7	55.7	56.2	55.7	55.4	54.2	53.6	52.9	56.2	62.5	64.1	72.5
Class12	48.9	46.9	47.8	39.7	51.2	46.5	53.1	59.6	73.3	55.9	52.5	60.0	55.7	61.2	60.6	69.6	80.4	81.6
Class13	97.9	95.9	96.6	97.2	96.6	95.2	96.5	99.3	98.6	98.6	98.6	97.9	98.6	97.9	98.6	96.5	99.3	98.6
Class14	75.4	75.7	73.8	84.4	74.9	76.2	77.5	85.6	86.5	76.5	76.7	80.7	88.2	81.8	78.7	84.4	93.2	89.8
Class15	48.7	46.9	52.8	41.7	47.8	48.2	49.6	74.5	87.1	64.7	58.6	58.9	65.3	69.6	62.6	74.8	86.8	86.5
Class16	85.9	85.9	85.9	84.6	88.5	84.6	85.9	85.9	90.1	62.8	73.1	89.7	83.3	88.5	89.7	84.6	85.9	98.7
AA（%）	70.2	68.7	69.6	67.8	70.8	67.6	70.4	79.0	84.3	73.5	74.2	76.0	76.4	79.1	76.5	80.9	85.6	91.1
OA（%）	60.9	60.1	60.2	58.4	60.6	59.7	61.5	70.8	73.2	66.6	66.2	67.1	67.0	68.6	66.9	73.7	79.1	84.7
KA（%）	56.0	55.1	55.2	53.5	55.6	54.7	56.7	67.1	69.8	62.3	61.9	62.8	62.7	64.5	62.6	69.9	77.6	82.6

Table 2. Classification performance of Indian Pines dataset with various methods（%）

View in the Article

Dataset	PCA	LPP	LDMGI	SPA	MSME	LESR	LMSDL	LDGSPP
PaviaU	0.02	0.12	2.50	2.24	3.39	46.55	57.47	81.01
Indian Pines	0.03	0.23	2.92	9.62	10.72	101.11	143.38	168.23

Table 3. Running time of different algorithms (s)

Download Citation

Save the article for my favorites

Paper Information