Table 1 (a) List of experiments and results. (b) List of methods and results. These experiments were performed with the specimens used for the third experiment (excluding broken and misidentified specimens).

From: Development of a system for the automated identification of herbarium specimens with high accuracy

(a)

Experiment No

Experiment name

Accuracy

Macro

Weighted

Number of herbarium images

Number of training images

Number of test images

Number of plant taxa

Number of family

Top1 (%)

Top5 (%)

Recall

Precision

f-score

Recall

Precision

f-score

1

All images

92

98

0.844

0.824

0.825

0.927

0.923

0.923

5,46,184

4,08,701

1,55,034

3114

219

2

w/o below 50 images species

94

99

0.898

0.892

0.892

0.942

0.93

0.939

5,34,778

3,85,536

1,49,242

2191

192

3

w/o broken or misidentified specimen

96

99

0.929

0.921

0.923

0.964

0.962

0.962

5,00,554

3,63,071

1,37,483

2171

192

4

w/o color-bar, stamp, scale

96

99

0.921

0.912

0.913

0.958

0.957

0.956

5,00,554

3,63,071

1,37,483

2171

192

5

Only Pteridophytes

98

100

0.946

0.947

0.945

0.985

0.984

0.984

2,04,174

1,26,218

77,956

357

32

(b)

 

Analysis method

Accuracy

Macro

Weighted

Number of herbarium images

Number of training images

Number of test images

Number of plant taxa

Number of family

Top1 (%)

Top5 (%)

Recall

Precision

f-score

Recall

Precision

f-score

 

Inception-ResNet-v2

96

99

0.929

0.921

0.923

0.964

0.962

0.962

5,00,554

3,63,071

1,37,483

2171

192

Inception-ResNet v2_base

95

99

0.913

0.905

0.906

0.955

0.953

0.953

Inception v3

95

99

0.909

0.901

0.902

0.953

0.951

0.951

VGG16

91

98

0.843

0.828

0.829

0.918

0.913

0.913