Table 2 Classification accuracy (%) at 28 dpi on leaf scans of three trials of Cassava-TME204-UCBSV and one trial of Cassava-Kiroba-UCBSV.

From: Early detection of plant virus infection using multispectral imaging and spatial–spectral machine learning

Groups

Methods

TME204

Kiroba

Trial 1

Trial 2

Trial 2 w PCR

Trial 3

 

Control versus infected

Vegetation indices

56.0  ±  13.1

87.7 ± 10.0

89.4 ±  9.3

73.9 ± 12.0

69.72 ± 21.22

Spectral reflectance (whole leaf)

67.6 ± 25.0

97.6 ±  8.3

96.9 ±  9.8

88.8 ± 16.4

87.41 ± 18.43

Spectral reflectance (patch-based)

78.9 ± 13.7

95.8 ±  5.8

96.1 ±  5.0

90.8 ± 11.3

87.98 ± 12.49

Spatia–spectral fusion (patch-based)

67.7 ± 13.6

94.1 ±  7.3

95.9 ±  6.3

89.1 ± 11.9

82.58 ± 14.50

Decision fusion (average)

76.9 ± 13.8

95.1 ±  6.6

95.6 ±  6.1

90.3 ± 11.5

83.67 ± 16.38

Spectral reflectance (patch-based voting)

87.2 ± 17.5

98.7 ±  5.6

98.2 ±  6.4

93.5 ± 14.5

90.60 ± 15.74

Spatial–spectral fusion (patch-based voting)

78.6 ± 20.9

98.3 ±  6.3

98.0 ±  5.8

91.7 ± 15.7

85.19 ± 20.07

ProbDecFus (patch-based)

79.0 ± 13.6

95.7 ±  6.3

96.5 ±  5.8

90.8 ± 11.4

88.12 ± 12.83

ProbDecFus (patch-based voting)

87.3 ± 17.4

98.5 ±  6.9

98.6 ±  5.1

93.6 ± 14.8

91.10 ± 15.74

Mock versus infected

Vegetation indices

53.6 ± 16.2

57.0 ± 14.9

56.3 ± 14.8

56.6 ± 17.5

63.21 ± 23.57

Spectral reflectance (whole leaf)

52.1 ± 21.8

75.3 ± 26.3

67.3 ± 22.3

76.1 ± 27.0

72.35 ± 28.27

Spectral reflectance (patch-based)

65.7 ± 13.3

72.6 ± 19.2

67.7 ± 17.6

76.9 ± 18.4

66.45 ± 22.80

Spatial–spectral fusion (patch-based)

52.8 ± 13.3

71.9 ± 17.3

66.9 ± 17.7

74.7 ± 16.5

69.66 ± 22.29

Decision fusion (average)

63.8 ± 15.0

70.9 ± 17.9

66.0 ± 16.9

74.2 ± 19.9

67.70 ± 24.84

Spectral reflectance (patch-based voting)

75.5 ± 20.8

75.4 ± 24.7

70.9 ± 24.1

81.0 ± 22.8

69.42 ± 27.54

Spatia–spectral fusion (patch-based voting)

57.1 ± 25.8

78.4 ± 25.6

71.1 ± 24.6

81.3 ± 21.8

76.37 ± 27.36

ProbDecFus (patch-based)

66.0 ± 13.3

72.3 ± 18.7

67.3 ± 17.4

77.6 ± 18.2

66.97 ± 23.30

ProbDecFus (patch-based voting)

75.9 ± 20.8

75.1 ± 24.5

70.2 ± 23.7

81.8 ± 22.6

69.47 ± 27.72

Control versus mock

Vegetation indices

55.3 ± 16.9

74.6 ± 15.5

N/A

67.8 ± 14.7

50.89 ± 16.45

Spectral reflectance (whole leaf)

60.8 ± 27.7

88.6 ± 18.9

79.4 ± 25.6

90.54 ± 15.36

Spectral reflectance (patch-based)

63.6 ± 17.7

91.4 ±  9.9

76.3 ± 15.5

84.56 ± 14.1

Spatial–spectral fusion (patch-based)

59.8 ± 15.5

90.4 ± 11.0

77.1 ± 15.2

81.80 ± 13.29

Decision fusion (average)

62.0 ± 17.9

88.5 ± 12.3

75.9 ± 15.3

80.11 ± 15.17

Spectral reflectance (patch-based voting)

70.7 ± 25.5

93.4 ± 11.5

84.3 ± 20.2

88.12 ± 19.01

Spatial–spectral fusion (patch-based voting)

66.9 ± 26.2

93.3 ±  14.0

84.7 ± 19.7

87.10 ± 20.52

ProbDecFus (patch-based)

63.2 ± 17.7

91.4 ± 10.4

76.5 ± 15.4

84.95 ± 14.22

ProbDecFus (patch-based voting)

79.0 ± 25.6

93.1 ± 11.9

84.5 ± 19.9

88.19 ± 19.23

  1. ‘Trial 2 w PCR’ of Cassava-TME204-UCBSV denotes the classification results of the models re-trained only on those leaves that were later confirmed of infection with PCR at 88 dpi.