Table 1 Experiment results of the U-Net model.

From: Eye in outer space: satellite imageries of container ports can predict world stock returns

Input size

Number of convolution layers

Accuracy

Recall

F-score

120*120

8

0.717

0.809

0.760

13

0.713

0.828

0.766

18

0.714

0.821

0.763

23

0.718

0.871

0.787

240*240

8

0.785

0.847

0.814

13

0.791

0.879

0.832

18

0.787

0.860

0.821

23

0.780

0.840

0.808

480*480

8

0.930

0.958

0.944

13

0.952

0.956

0.954

18

0.959

0.948

0.953

23

0.955

0.961

0.958

600*600

8

0.848

0.863

0.855

13

0.841

0.841

0.840

18

0.839

0.859

0.848

23

0.831

0.883

0.855

  1. This table compares the model performance under different groups of hyperparameters, including the input size of images and the number of convolution layers. The accuracy, recall, and F-score are evaluated through 10-fold cross-validation.
  2. Bold values represents the maximum value achieved under the corresponding criterion.
Back to article page