Table 4 Quantitative comparison between the proposed method and nine alternative approaches

From: Multi-objective band selection algorithm based on NSGA-II for pattern segmentation of textile hyperspectral images

Method

PA (%)

IoU (%)

Precision (%)

Recall (%)

F1 (%)

Time(s)

Our method

93.74

73.19

75.78

95.54

84.52

5.73

Gaussian filter method on HSI

80.65

45.34

73.07

54.01

62.17

5.36

HSI single band method

76.57

48.23

54.23

81.34

65.07

1.06

Grayscale image method on RGB

77.53

42.46

57.57

61.80

59.61

1.16

Vector method on RGB

79.54

49.48

67.04

66.11

66.57

1.51

Vector method by using all channels on HSI

92.00

73.09

73.10

92.98

80.76

5.42

Euclidean distance method on HSI

79.61

64.87

66.78

95.36

78.69

6.44

SAM distance method on HSI

80.46

52.96

59.92

82.01

69.25

10.69

Traditional NMS method on HSI

82.13

46.66

70.10

58.25

63.63

4.72

Manual threshold method on HSI

81.55

47.20

59.02

66.96

62.77

24.71

  1. Bold values denote the optimal performance metrics among the compared methods in the respective evaluation tasks.