Facial occlusion emotion recognition method based on improved attention mechanism

The invention discloses a facial occlusion emotion recognition method based on an improved attention mechanism. Belongs to the field of image processing, and aims to realize facial emotion recognition under the shielding condition so as to reduce the influence of difficulty in emotion recognition and poor recognition effect caused by facial shielding. The method comprises the following steps: 1, preprocessing an original image, and inputting the preprocessed original image into a feature extraction network A with ResNet18 as a main network to obtain global features and shallow features; 2, performing segmentation processing on the shallow features to obtain four local feature maps, and respectively extracting local features by adding an attention-improving feature extraction network B; 3, flattening the obtained global features and local features into one-dimensional feature vectors; step 4, the flattened feature vectors are sent to a feature fusion network C which takes Encoder of Transformer as a main part