Face expression recognition based on feature fusion and attention mecha-nism
This paper addresses the issue of insufficient focus on facial expression features and in-adequate feature extraction leading to the loss of valuable information in existing single-channel convolutional neural networks for facial expression recognition.A facial expression recognition algorithm is proposed,which integrates features from multiple channels through parallel net-works.The algorithm combines local detailed features with global holistic features and employs dual networks to extract features from different channels,achieving a combination of coarse and fine granularity to enhance the model's recognition capability for various expressions.The algo-rithm utilizes dual feature extraction channels,including the M-CNN channel,which uses a re-sidual network with attention mechanisms to extract global features,and the P-CNN channel,which employs a multi-scale feature extraction network to extract eye and mouth features.Sub-sequently,the features extracted from the three channels are fused,dimensionality is reduced af-ter channel importance is determined through a channel attention module,and the fused features are finally fed into a joint loss function layer for classification.The model has undergone exten-sive experiments on the CK+dataset and FER2013 dataset.The results demonstrate an improve-ment in recognition accuracy compared to other state-of-the-art methods,validating the advance-ment and effectiveness of the proposed algorithm.
Deep learningFacial expression recognitionFeature fusionAttention mechanism
万方数据
维普
