Table 3 Quantitative comparisons of the BDCI dataset with other models.

From: A deep learning method for optimizing semantic segmentation accuracy of remote sensing images based on improved UNet

Method

Per-class F1 score

Avg. F1

OA

mIOU

Woodland

Cultivated field

Grassland

Building

Road

Water

Unet

90.14

92.28

18.82

79.61

40.82

92.54

69.04

88.59

62.36

FCN + SE

90.68

92.7

19.59

93.94

36.06

93.63

71.1

89.74

62.18

DeepLabV3 + 

89.62

92.72

31.17

84.86

39.38

93.33

71.85

89.53

65.47

PSPnet

90.18

92.96

28.92

84.28

36.11

93.24

70.95

89.99

65.36

DAnet

90.79

93.18

28.35

84.65

47.7

94.28

73.16

90.33

67.31

CBAM

90.81

93.21

31.02

85.66

47.98

94.33

73.84

90.33

67.89

Proposed

91.96

94.05

37.83

86.45

52.66

95.09

76.34

91.65

70.5

  1. Significant values are in bold.