Skip to main content
. 2021 Jul 1;21(13):4536. doi: 10.3390/s21134536
Algorithm 1 CNN-LRP
Input: the original SAR image I in size of (1,W,H)
  • 1:

    model f

  • 2:

    parameters: α, β

  • 3:

    forl in [ L−1,...,1] do:

  • 4:

        if l in classification layers then:

  • 5:

                 zn,m as in Equation (2)

  • 6:

                 rn(l) as in Equation (5)

  • 7:

                 Rl as in Equation (4)

  • 8:

        else if l in convolution layers then:

  • 9:

                 z[n,c,w,h] as in Equation (6)

  • 10:

                 rn(l) as in Equation (5)

  • 11:

                 Rl as in Equation (4)

  • 12:

        end if

  • 13:

        if l in maxpooling layers then:

  • 14:

                  RQl as in Equation (7)

  • 15:

        end if

  • 16:

    end for

  • 17:

    H(LRP) = R(1)

Output: the heatmap H(LRP) in size of (1,W,H)