Table 4 The comparative experimental results based on different CNNs.

From: Visual feature-based multi-scale hybrid attention network for fine-grained Hawthorn varieties identification

CNN module

Top-1 ACC (%)

Params (M)

FLOPs

CNN block

86.43

31.42

10.51

Residual block

87.27

56.34

4.57

Dense block

88.65

89.97

4.18

Depthwise Separable block

89.82

10.83

3.94

Our

90.96

15.92

4.76

Back to article page