Skip to main content

Table 3 Experimental results of our feature reduction techniques between PCA, 2DCAE, and 3DCAE. An explanation of the metrics can be found in "Experiment configurations" section

From: CEU-Net: ensemble semantic segmentation of hyperspectral images using clustering

Methods

OA

AA

Kappa

OA

AA

Kappa

 

IP (k = 2)

Salinas (k = 3)

PCA

90.01 ± 0.1

90.52 ± 0.2

88.67 ± 0.1

96.44 ± 0.1

98.36 ± 0.1

96.34 ± 0.1

2D-CAE

65.39 ± 0.2

51.39 ± 0.2

60.04 ± 0.2

85.97 ± 0.2

91.65 ± 0.2

84.36 ± 0.2

3D-CAE

70.50 ± 0.2

55.84 ± 0.3

61.21 ± 0.2

87.02 ± 0.2

91.71 ± 0.2

85.94 ± 0.2

 

KSC (k = 2)

Botswana (k = 3)

PCA

95.25 ± 0.1

93.05 ± 0.1

94.98 ± 0.1

96.43 ± 0.2

97.1 ± 0.2

96.13 ± 0.2

2D-CAE

90.02 ± 0.2

84.39 ± 0.2

88.87 ± 0.2

91.26 ± 0.2

91.84 ± 0.2

90.52 ± 0.2

3D-CAE

91.10 ± 0.2

85.62 ± 0.2

89.46 ± 0.2

93.12 ± 0.2

93.44 ± 0.2

91.45 ± 0.2

 

PU (k=2)

Houston (k=2)

PCA

96.18 ± 0.1

95.10 ± 0.1

95.00 ± 0.1

98.49 ± 0.1

98.38 ± 0.1

98.36 ± 0.1

2D-CAE

80.94 ± 0.2

75.85 ± 0.2

73.85 ± 0.1

51.57 ± 0.4

54.00 ± 0.4

47.76 ± 0.4

3D-CAE

81.47 ± 0.2

78.45 ± 0.1

74.99 ± 0.2

74.35 ± 0.3

73.81 ± 0.4

72.25 ± 0.3

  1. Highest performing values are highlighted in bold