Table 8 Comparison of different feature fusion strategies.

From: A novel approach to palm vein image segmentation combining multi-scale convolution and swin-transformer networks

Fusion strategy

OA

Dice

IoU (%)

Parameter increment

FLOPs

Channel concatenation

97.8%

94.5%

89.4%

+0%

32

Weighted average

96.9%

93.7%

88.1%

+0.1%

35

Channel attention weighting

97.5%

94.2%

89.0%

+0.3%

30

Spatial attention weighting

97.3%

93.9%

88.6%

+0.5%

28