Answers are grounded in this paper and cite their sources.
When using convolutional neural networks (CNNs) for facial emotion recognition, down-sampling in the convolutional layers often leads to the loss of many features. These lost features also play a role in emotion recognition and can affect the effectiveness of emotion recognition. Therefore, we propose a facial emotion recognition method based on auxiliary classifiers and SENet modules to effectively alleviate the problem of feature loss during the convolution process. By using SENet modules, the model enhances its focus on different channel features during the feature extraction process, enabling the model to pay better attention to important features. Subsequently, auxiliary classifiers are introduced into the model to effectively utilize features extracted from different layers for auxiliary classification. Finally, the auxiliary classification results are adaptively weighted and fused with the main classifier's classification results. Experimental results demonstrate that introducing auxiliary classifiers and SENet modules further improves the model's accuracy, reaching 68.87%.
No takes yet. Share an insight, caveat, or question.
Shang et al. (2024) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: