Chen
et al.
2011; Tropp and Gilbert 2007), simultaneous or-
thogonal matching pursuit (Y. Chen
et al.
2011), sparse coding
(Lee
et al.
2007) and collaborative sparse coding. To compare
the performance of these methods, the parameters were set
according to previous studies (Y. Chen
et al.
2011; Li
et al.
2014; W. Li
et al.
2015). For fair comparison, all methods used
the same training and testing data after preprocessing with
MH
prediction.
Experimental Results: Indian Pines
The first
HSI
was collected over the northwest of Indiana’s Indian
Pines test site in June 1992. The
HSI
has 220 bands with 145×145
pixels in the region of the visible and infrared spectrum from
0.4 to 2.45
μ
m with a spatial resolution of 20 m. The water-
absorption bands were removed, resulting in 200 spectral bands.
The original Indian Pines data set consists of samples from 16
land cover classes with ground-truth data. Some of the classes
have small numbers of samples, hence increasing the difficulty
of classification, especially for rare classes.
The numbers of training and testing samples from each
class are shown in Table 1. The training and testing sets are
visually shown in Figure 11a and 11b, respectively. The
classification accuracy for each class, the overall accuracy,
the average accuracy, and the kappa coefficient measure are
shown in Table 2. The classification maps for the labeled
pixels are shown in Figure 11. An image-patch size
ws
= 3×3
Figure 11. The Indian Pines hyperspectral-image classification results: (a) training
set and (b) testing set. Classification maps obtained by (c) joint collaborative
representation, (d) kernel collaborative representation with Tikhonov regularization
and composite kernel, (e) support vector machine, (f) simultaneous orthogonal
matching pursuit, (g) orthogonal matching pursuit, (h) nearest regularized subspace, (i)
sparse coding, (j) collaborative representation, (k) collaborative sparse coding, and (l)
collaborative sparse coding with smoothness regularization.
Table 1. Listing of 16 classes in the
Indian Pines hyperspectral-image
data. Training and testing sample
counts are shown for each class.
Class
Samples
No. Name
Train Test
1 Alfalfa
5
41
2 Corn-notill
143 1285
3 Corn-min
83 747
4 Corn
24 213
5 Grass/Pasture
49 434
6 Grass/Tree
73 657
7 Grass/Pasture-
mowed
3
25
8 Hay-windrowed
48 430
9 Oats
2
18
10 Soybeans-notill
98 874
11 Soybeans-min
246 2209
12 Soybeans-clean
60 533
13 Wheat
21 18
14 Woods
127 113
15 Building-Grass-
Trees-Drives
39 34
16 Stone-steel
Towers
10
83
Total
1031 9218
Table 2. Classification accuracy (%) for the testing set of the Indian Pines image. Boldface indicates the best-performing classifier(s) for
each class.
Notes
: SVM = support vector machine; OMP = orthogonal matching pursuit; CRC = collaborative representation; SOMP =
simultaneous orthogonal matching pursuit; NRS = nearest regularized subspace; JCR = joint collaborative representation; KCRT-CK =
kernel collaborative representation with Tikhonov regularization and composite kernel; SC = sparse coding; CSC = collaborative sparse
coding; CSCSR = collaborative sparse coding with smoothness regularization; OA = overall accuracy; AA = average accuracy.
Class
SVM OMP
CRC SOMP NRS
JCR KCRT-CK
SC
CSC CSCSR
1
39.02
97.56
100
92.68
70.73
73.17
97.56
92.68
100
100
2
76.73
90.43
96.89
92.22
91.44
93.23
96.58
80.08
88.56
97.82
3
81.53
91.17
97.05
93.44
93.71
95.98
95.18
79.92
93.04
98.80
4
70.89
84.98
97.65
91.08
69.01
72.30
100
77.93
92.49
100
5
91.71
97.005
90.55
96.77
92.63
96.31
98.62
89.40
95.39
99.31
6
100
100
96.65
98.78
99.24
100
100
96.50
99.54
99.54
7
88.00
36.00
96.00
100
52.00
36.00
92.00
92.00
100
100
8
99.07
100
100
100
99.77
100
99.77
98.61
99.77
100
9
27.78
11.11
55.56
100
33.33
5.56
83.33
66.67
66.67
83.33
10
83.98
72.77
99.08
91.08
90.39
93.82
98.86
87.41
96.34
100
11
94.7
98.78
99.14
96.70
99.55
99.68
99.46
90.18
96.33
99.91
12
77.67
89.87
97.37
90.62
84.62
92.68
97.00
70.36
86.68
98.12
13
99.46
98.37
99.46
98.91
97.83
98.91
99.46
99.46
100
99.46
14
93.23
99.82
100
99.30
98.77
99.38
99.65
92.09
95.78
100
15
89.34
100
100
100
95.39
97.41
99.14
72.62
89.63
100
16
84.34
98.79
97.59
81.93
89.16
98.80
93.98
100
100
OA
88.24
93.59
97.92
95.60
94.34
96.06
98.48
86.82
94.44
99.39
AA
81.09
85.49
92.14
96.20
84.40
83.97
97.21
86.24
93.76
98.55
κ
86.55
92.66
97.64
94.98
93.58
95.53
98.27
84.98
93.65
99.31
666
September 2019
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
