Multichannel Retinal Blood Vessel Segmentation Based on the Combination of Matched Filter and U-Net Network

Algorithm 1

Training of U-Net with dynamic learning rate.

Input: Train images, ground truth

Input: Initial epochs 30, batch size 1, learning rate 0.01

Input: Initialize best loss Inf

Output: Predicted images , U-Net parameter

1. preprocessing ()

2. enhancement ()

3. for 0 to do

4. ifthen

5.   

6.   else ifthen

7.     

8. end if

9.  compute the number of train images

10.  initial parameter of U-Net

11. whiledo

12.    ()

13.   

14.   ifthen

15.     

16.     

17.   end if

18.   

19. end while

20. end for

21.

22. return,