Fig 1.
The overall network architecture of HFE-Net.
Fig 2.
EfficientNetV2 [42] network architecture.
Fig 3.
(a) shows the working principle of self-attention mechanism, (b) shows the working principle of Multi-head Self-attention mechanism.
Table 1.
Distribution of data in the FER2013 dataset.
Table 2.
Distribution of data in the FERPlus dataset.
Table 3.
Distribution of data in the RAF-DB dataset.
Table 4.
Distribution of data in the Affectnet dataset.
Table 5.
Multilayer perceptron, Multi-head Self-attention and Feature Fusion Device ablation experiments on four FER datasets.
Table 6.
Comparison of experimental results on the FER2013 dataset.
Table 7.
Comparison of experimental results on the FERPlus dataset.
Table 8.
Comparison of experimental results on the RAF-DB dataset.
Table 9.
Comparison of experimental results on the Affectnet dataset.
Fig 4.
Confusion matrix of Effv2-s (baseline) on three datasets.
Fig 5.
Confusion matrix of the proposed model on three datasets.