Hyperspectral Fast Spectral Clustering Algorithm Based on Multi-Layer Bipartite Graph

Siyuan Li; Zhiyuan Zheng; Xiaoyan Du; Tong Liu; Xiaojun Yang

doi:10.3788/LOP202259.1210009

Journals >Laser & Optoelectronics Progress >Volume 59 >Issue 12 >Page 1210009 > Article

Laser & Optoelectronics Progress
Vol. 59, Issue 12, 1210009 (2022)

Hyperspectral Fast Spectral Clustering Algorithm Based on Multi-Layer Bipartite Graph

Siyuan Li¹, Zhiyuan Zheng², Xiaoyan Du³, Tong Liu^1、*, and Xiaojun Yang¹

Author Affiliations

¹College of Information Engineering, Guangdong University of Technology, Guangzhou 510006, Guangdong , China

²Rocket Force University of Engineering, Chinese People’s Liberation Army, Xi’an 710025, Shaanxi , China

³Chinese People’s Liberation Army 96962 Troops, Beijing 102206, China

show less

DOI: 10.3788/LOP202259.1210009 Cite this Article Set citation alerts

Siyuan Li, Zhiyuan Zheng, Xiaoyan Du, Tong Liu, Xiaojun Yang. Hyperspectral Fast Spectral Clustering Algorithm Based on Multi-Layer Bipartite Graph[J]. Laser & Optoelectronics Progress, 2022, 59(12): 1210009 Copy Citation Text

show less

Fig. 1. Multi-layer anchor graph

Download full size

Clustering graphs of different algorithms in Indian Pines data set. (a) Real image; (b) K-means algorithm; (c) FCM algorithm; (d) FCM_S1 algorithm; (e) SC algorithm; (f) FSCBG-128 algorithm; (g) FCMBG-512-128 algorithm

Fig. 2. Clustering graphs of different algorithms in Indian Pines data set. (a) Real image; (b) K-means algorithm; (c) FCM algorithm; (d) FCM_S1 algorithm; (e) SC algorithm; (f) FSCBG-128 algorithm; (g) FCMBG-512-128 algorithm

Download full size

Fig. 3. Clustering graphs of different algorithms in Salinas data set. (a) Real image; (b) K-means algorithm; (c) FCM algorithm; (d) FCM_S1 algorithm; (e) FSCBG-256 algorithm; (f) FCMBG-1024-256 algorithm

Download full size

Input：Original data X，the number of classes c，the number of anchors in each layer m₁，...，m_h，the number of near neighbor k

1) Generate m anchors by balanced K‑means based hierarchical K‑means selection;

2) Obtain the matrix $Z_{0,1}$ ，…， $Z_{h - 1, h}$ ，according to Eq. (4);

3) Obtain the matrix $Z_{H}$ according to Eq. (3);

4) Calculate the adjacent matrix of bipartite hierarchical graph W= $[\begin{matrix} 0 & Z_{H} \\ Z {}_{H}^{T} & 0 \end{matrix}]$ ;

5) Obtain the relaxed continuous solution of F_X and F_U by performing singular value decomposition on matrix B= $Z_{H} Λ^{- \frac{1}{2}}$ .

Output：c classes

Table 1. Procedure of FCMBG algorithm

Data set	Data volume	Dimension	Class number
Indian Pines	21025	220	16
Salinas	111104	204	16

Table 2. Details of hyperspectral data sets

Algorithm	K-means	FCM	FCM_S1	SC	FSCBG-128	FCMBG-512-128
AA /%	36.57	36.83	34.76	31.12	35.25	39.89
OA /%	34.84	35.42	36.58	33.16	35.08	37.80
Kappa	0.2758	0.2859	0.2953	0.2725	0.2866	0.3022
Time /s	3.8601	4.2935	6.2047	39.2537	0.7431	1.3250

Table 3. Experimental results of different algorithms in Indian Pines data set

Algorithm	K-means	FCM	FCM_S1	FSCBG-256	FCMBG-1024-256
AA /%	64.93	65.79	66.05	64.11	68.41
OA /%	66.02	64.42	66.41	64.90	70.13
Kappa	0.6216	0.6062	0.6264	0.6128	0.6684
Time /s	20.1429	21.5936	35.2432	3.1920	11.4740

Table 4. Results of Salinas data set

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information