Table 1 Statistical results of the three kinds of classification experiments, performed on the set of 131 drainage networks.

From: A novel approach to the classification of terrestrial drainage networks based on deep learning and preliminary results on solar system bodies

Best Q-test

Data type

Train set

Test set

Cleaned data

D: 45

ND: 58

D: 13

ND: 15

Statistics on test set

Model

Class

P

C

F1

UC

VGGNet + Adadelta

D

100

100

100

–

ND

100

100

100

–

VGGNet + RF

D

92

92

92

Turkey

ND

93

93

93

Devoll

AlexNet + Adadelta

D

100

77

87

Tambun Varuna Vishwamitri

ND

83

100

91

–

Best Rob-test

Data type

Train set

Test set

Noisy data

D: 45

ND: 58

D: 13

ND: 15

Statistics on test set

Model

Class

P

C

F1

UC

VGGNet + Adadelta

D

100

57

73

Po, Tambun, Varuna

ND

77

100

87

–

VGGNet + RF

D

80

62

70

Candelaro, Po, Tambun, Varuna, Vishwamitri

ND

72

87

79

Chari, Devoll

AlexNet + Adadelta

D

92

85

88

Tambun, Varuna

ND

88

93

90

Devoll

Best Rel-test

Data type

Train set

Test set

Cleaned data

D: 53

ND: 50

D: 13

ND: 15

Statistics on test set

Model

Class

P

C

F1

UC

VGGNet + Adadelta

D

79

85

82

Po, Vishwamitri

ND

86

80

83

Chari, Devoll, Pamba

VGGNet + RF

D

85

85

85

Candelaro, Vishwamitri

ND

87

87

87

Chari, Pamba

AlexNet + Adadelta

D

79

85

82

Po, Vishwamitri

ND

86

80

83

Chari, Devoll, Pamba

  1. P purity, C completeness, and F1 F1-score, are in percent. UC reports the list of mismatched samples. D dendritic, ND non-dendritic pattern.