IEEE Geoscience and Remote Sensing Magazine - September 2021 - 59

different methods. The experiments with four-connectivity
provided better scores in general. We attribute this result to
the larger number of tree nodes obtained using four- instead
of eight-connectivity. Hence, we conclude that the finer regions
obtained with four-connectivity provide valuable attributes
for performing classification on this data set.
Regarding the Pavia data, changing the connectivity pa22102
22102
11101
00011
20012
rameter
had a different impact on each method. All methods
perform better on the four-connected Pavia data, except
for
FP .a Notably, changing the connectivity parameter
of the AP-minT from eight to four led to improvements of
8.89% and 11.02% in terms of OA and
l , respectively. On
the other hand, changing the connectivity parameter of the
FPa
from four to eight led to improvements of 4.09% and
4.72% in terms of OA and l , respectively.
Overall, though, four-connectivity can be observed to
almost consistently outperform eight-connectivity. In terms
of future research directions, advanced connectivity concepts,
such as hyperconnectivity [104] and mask-based
connectivity [105], appear as promising options.
We now focus on the assessment of the quantization parameter
in the classification of the Pavia data set. Since the
(b)
(c)
FIGURE 7. (a) The original grayscale image :, ,.
XE 01 2 " 6
@ (b) The
max-tree of X computed using four-connectivity. (c) The max-tree
of X computed using eight-connectivity.
TABLE 10. A COMPARISON OF THE CLASSIFICATION RESULTS OF GRAY-POTSDAM FOR DIFFERENT CONNECTIVITY PARAMETERS.
METHOD
AP-maxT
AP-minT
AP
a-AP
~-AP
FPn
FPa
FPn + a
TF-AP
DIMENSIONS
20
20
38
20
20
38
38
76
18
OA (%)
CLASSIFICATION RESULT (FOUR-CONNECTIVITY)
AA (%)
l # 100
63.65 ! 0.14
58 ! 0.23
75.1 ! 0.05
68.32 ! 0.07
68.1 ! 0.06
77.14 ! 0.05
77.19 ! 0.05
77.94 ! 0.04
72.34 ! 0.05
59.68 ! 0.02
56.55 ! 0.02
77.56 ! 0.02
67.94 ! 0.04
67.78 ! 0.03
79.41 ! 0.02
79.46 ! 0.02
80.17 ! 0.03
74.93 ! 0.04
52.65 ! 0.15
47.13 ! 0.21
67.83 ! 0.05
59.47 ! 0.08
59.23 ! 0.06
70.42 ! 0.06
70.47 ! 0.05
71.43 ! 0.05
64.33 ! 0.06
OA (%)
CLASSIFICATION RESULT (EIGHT-CONNECTIVITY)
AA (%)
l # 100
63..
57..
73..
67..
67..
75..
75..
76..
71..
03 016!
36 023!
91 007!
67 008!
4008!
95 007!
98 009!
7006!
9008!
58..
55..
76..
66..
66..
78..
78..
78..
74..
9002!
75 003!
42 003!
72 004!
46 003!
23 002!
28 003!
96 003!
49 003!
The values in bold represent the highest scores when comparing the scores on the left (four connectivity) with the scores on the right (eight-connectivity).
51..
46..
66..
58..
58..
68..
68..
69..
63..
84 017!
3021!
36 007!
64 008!
34 008!
91 008!
96 01!
86 007!
81 009!
L1
L2
L1
L2
(a)
L0
L0
TABLE 11. A COMPARISON OF THE CLASSIFICATION RESULTS OF PAVIA FOR DIFFERENT CONNECTIVITY PARAMETERS.
METHOD
AP-maxT
AP-minT
AP
a-AP
~-AP
FPn
FPa
FPn + a
TF-AP
DIMENSIONS
80
80
152
80
80
152
152
304
72
OA (%)
CLASSIFICATION RESULT (FOUR-CONNECTIVITY)
AA (%)
l # 100
89.21 ! 0.63
87.11 ! 2
90.57 ! 2.6
95.46 ! 0.59
96.33 ! 0.4
96.76 ! 0.15
85 ..
48 098!
96.15 ! 0.14
93.98 ! 0.53
87..
37 019!
92.62 ! 0.61
93..
95..
13 067!
25 123!
97.17 ! 0.89
97.05 ! 0.19
93 ..
94 055!
96.74 ! 0.32
96.26 ! 0.29
85.42 ! 0.8
83.23 ! 2.41
87.53 ! 3.32
93.91 ! 0.79
95.08 ! 0.54
95.65 ! 0.21
81 ..
36 116!
94.82 ! 0.19
92.04 ! 0.68
OA (%)
87..
78..
90 ..
94..
94..
93..
88 116!
22 05!
54 155!
73 04!
71 034!
35 015!
89.57 ! 1.18
94..
92..
21 032!
65 024!
89.56 ! 0.3
88 ..
93.8 ! 0.4
26 014!
95.66 ! 0.27
95 ..
93 ..
69 017!
55 016!
94.03 ! 0.32
95 .028!
95 ..
26 017!
CLASSIFICATION RESULT (8-CONNECTIVITY)
AA (%)
l # 100
84..
72 ..
87..
92..
92 ..
90 ..
17 143!
21 057!
31 199!
97 052!
94 044!
902!
86.08 ! 1.5
92 ..
90 ..
The values in bold represent the highest scores when comparing the columns on the left (four-connectivity) with their corresponding columns on the right (four-connectivity).
SEPTEMBER 2021 IEEE GEOSCIENCE AND REMOTE SENSING MAGAZINE
59
09 044!
32 031!
IEEE Geoscience and Remote Sensing Magazine - September 2021

Table of Contents for the Digital Edition of IEEE Geoscience and Remote Sensing Magazine - September 2021
