An improved cuckoo search-based adaptive band selection for hyperspectral image classification

Figures & data

Figure 1. Proposed band selections system with two CSs

Table 1. Computational complexity of different algorithms

		Number of Multiplications
Searching method	1CS	O(nXC)
	2CS	O(niXiCinoXoCo )
	PSO	O(nXC)
	FA	O(nXC)
	SFFS	O(b)
	SFS	O(b)
Criterion function	JM	${\sum }_{k=1}^{b}O\left(k^3{p}^{2}L\right)+O\left(k^{2}sL\right)+O\left(k{p}^{2}L\right)$
	MEAC	${\sum }_{k=1}^{b}O\left(k^{3}L\right)+O\left(k^{2}pL\right)+O\left(k{p}^{2}L\right)$

-n denotes the number of initial nests; X the number of iterations; C the complexity of cost function; subscript i the inner CS, subscript o the outer CS, b the number of selected bands.

-L denotes the number of bands, p the number of classes, and s the number of total training samples.

Table 2. Training and testing samples for HYDICE data

Class	Name	Training No.	Test No.
1	Road	55	892
2	Grass	57	906
3	Shadow	50	587
4	Trail	46	578
5	Tree	49	630
6	Roof	69	1500
Total		309	4772

Figure 2. HYDICE DC MALL data

Table 3. Training and testing samples for HyMap Data

Class	Name	Training No.	Test No.
1	Road	73	1231
2	Grass	72	1072
3	Shadow	49	215
4	Soil	69	380
5	Tree	67	1321
6	Roof	74	1244
Total		404	5463

Figure 3. HyMap Purdue Campus data

Table 4. The overall classification accuracy for HYDICE data

	Overall Classification Accuracy (OA)(%)
	Minimum	Maximum	Average
2CS- MEAC	95.1	97.2	96.2
1CS-MEAC (6 bands)	91.6	94.8	93.0
2CS-JM	95.6	96.9	96.3
1CS-JM (6 bands)	93.6	95.4	94.2
PSO-MEAC(6 bands)	89.2	93.5	92.0
PSO-JM(6 bands)	92.4	94.0	93.3
FA-MEAC(6 bands)	92.3	94.1	93.5
FA-JM(6 bands)	92.0	94.5	93.4
SFS-MEAC (6 bands)	92.7
SFS-JM (6 bands)	89.2
SFFS-MEAC (6 bands)	92.7
SFFS-JM (6 bands)	89.2
PCA (6PCs)	92.7
All bands	93.4

Figure 4. Classification map of HYDICE data using MEAC and JM.(a) 2CS-MEAC (97.2%). Classification map of HYDICE data using MEAC and JM.(f)All bands (93.4%)

Table 5. The overall classification accuracy for HyMap data

	Overall Classification Accuracy (OA) (%)
	Minimum	Maximum	Average
2CS- MEAC	94.3	95.8	94.6
1CS-MEAC (13 bands)	90.1	94.3	92.7
2CS-JM	93.2	96.2	94.6
1CS-JM (13 bands)	92.6	94.4	93.7
PSO-MEAC(13 bands)	88.2	93.0	91.1
PSO-JM(13 bands)	91.8	93.4	92.6
FA-MEAC(13 bands)	91.5	93.5	92.3
FA-JM(13 bands)	92.0	94.1	93.2
SFS-MEAC (13 bands)	81.9%
SFS-JM (13 bands)	88.1%
SFFS-MEAC (13 bands)	82.0%
SFFS-JM (13 bands)	90.2%
PCA (13 PCs)	92.7%
All bands	88.7%

Figure 5. Classification map of HyMap data using MEAC and JM.(a) 2CS-MEAC (95.8%). Classification map of HyMap data using MEAC and JM.(b) 1CS-MEAC (94.3%). Classification map of HyMap data using MEAC and JM.(c (c) 2CS-JM (96.2%). Classification map of HyMap data using MEAC and JM.(d) 1CS-JM (94.4%). Classification map of HyMap data using MEAC and JM.(e) PCA (92.7%). Classification map of HyMap data using MEAC and JM.(f) All bands (88.7%)

Table 6. Computing time of different algorithms

	Time (in second)
HYDICE DC Mall	1CS	5.7
	2CS	2131.2
	PSO	5.6
	FA	5.5
	SFFS	8.1
	SFS	5.5
HyMap Purdue	1CS	8.6
	2CS	2467.5
	PSO	8.3
	FA	8.0
	SFFS	4.2
	SFS	3.8