Facial expression recognition based on multi-scale dual-channel attention mechanism

In recent years, with the continuous advancement of artificial intelligence technology, the way humans interact with computers has undergone significant changes. Facial expression recognition technology for human faces has increasingly become a hot research topic in computer vision. However, traditional facial expression recognition models are based on deep neural networks, which have problems such as insufficient extraction of image details, poor generalization ability of the model, and low utilization rate of key feature information. To address these issues, in this paper, a Multi-scale Dual-channel Attention Mechanism Facial Expression Recognition Network (MDAMNet) is proposed. Firstly, a Multi-scale Feature Extraction Block with Dilated Convolution (DMFE Block) is designed, which uses ordinary convolution and dilated convolution to form different receptive fields. The model can focus on and extract feature information at different scales at the same time. Then, a Dual-channel Attention Mechanism (DAM) combining channel attention and spatial attention is designed, which can enhance the perception ability and extraction ability of key features, and effectively suppress redundant information. Finally, Focal Loss is employed to dynamically adjust the loss function, with the aim of enhancing the accuracy and stability of the system. During the experimental verification stage, experiments were conducted on three publicly available classic datasets (FER2013, CK + and JAFFE). The final accuracy rates reached 76.87%, 97.52% and 98.34% respectively, effectively demonstrating the feasibility of this method.