Skip to main content
. 2026 Jun 16;12(6):263. doi: 10.3390/jimaging12060263
Algorithm 2 FGCA-YeastNet for Multimodal Yeast Classification

  • Require: Microscopy image Im, light-scattering image Is

  • Ensure: Predicted yeast category y

  • 1:

    Begin

  • 2:

    Extract modality-specific features:

  • 3:

       Fs ← FPA(Stem(Is))

  • 4:

       Fm ← Stem(Im)

  • 7:

    Pass features through stacked convolution blocks:

  • 8:

       Fs, Fm ← ConvBlock(Fs, Fm)

  • 6:

    Perform bidirectional coarse alignment using BiCAM:

  • 7:

       Fˆm, Fˆs ← BiCAM(Fm, Fs)

  • 8:

    Apply frequency-guided cross-attention:

  • 12:

     F˜m, F˜s ← FGCA(Fˆm, Fˆs)

  • 12:

    Refine features using convolution blocks:

  • 13:

     Gs ← ConvBlock(F˜s)

  • 14:

     Gm ← ConvBlock(F˜m)

  • 15:

    Perform global average pooling:

  • 16:

     Ps ← AvgPool(Gs)

  • 17:

     Pm ← AvgPool(Gm)

  • 18:

    Concatenate pooled features:

  • 19:

     Ffusion ← Concat(Ps, Pm)

  • 12:

    Predict yeast category using classifier:

  • 13:

     y ← Classifier(Ffusion)

  • 14:

    Return y

  • 15:

    End