Table 3 Performance of deep learning models

From: A detection-screening framework for karez (ancient underground irrigation system) using deep learning and geospatial analysis

Model type

Backbone model

Epochs

Precision

Training time (Min)

Average time for each Epoch

MMDetection

ResNet34

29

0.810

1163

40.10

MMDetection

ResNet50

31

0.812

1179

38.03

SSD

ResNet34

120

0.711

494

4.12

SSD

ResNet50

200

0.751

1402

7.01

YoloV3

DarkNet53

27

0.717

253

9.37

Faster R-CNN

ResNet34

27

0.442

454

16.82

Faster R-CNN

ResNet50

27

0.773

407

15.07

  1. The models were trained by setting “stop training when models stop improving”. The maximum epochs were set 200, and most models were stopped earlier because of no improvement.
Back to article page