Research on Occluded Facial Expression Recognition Algorithm Based on Hybrid Attention Mechanism and Multi-scale Feature Extraction

In practical applications, the recognition performance of classical convolutional neural networks is often greatly reduced when faced with problems such as uneven facial lighting, changes in head posture, and local occlusion in real environments. Therefore, this paper proposes a network model based on hybrid attention mechanism and multi-scale feature extraction. First, ResNet-18 is used to extract the preliminary features of face images. ResNet-18 has good depth and structural stability, and can extract relatively basic facial features. Then, global features are extracted through multi-scale modules. This multi-scale convolution operation can capture the information of face images from different angles and scales, which greatly increases the generalization ability of the network model and enables it to adapt to more different situations of face images. At the same time, the hybrid attention mechanism can extract the local features of the expression image. The attention mechanism can make the model focus more on the key areas related to the expression in the image, ignore some irrelevant or interfering information, and thus increase the robustness of the network model. Finally, through the feature fusion module, the extracted global features and local features are fused, so that the model can comprehensively and accurately understand facial expression information, and output accurate classification results. Experimental results show that the algorithm can effectively capture the local features and global information of facial expressions, and has achieved remarkable results in the accuracy and robustness of facial expression recognition.