Expression recognition method in natural scene based on brain-like technology

The invention provides an expression feature extraction model, an expression recognition method in a natural scene based on the expression feature extraction model, and an expression recognition method in the natural scene based on a brain-like technology. According to the expression feature extraction model, an SCAN model is improved, wherein backbone ResNet50 is replaced with pre-trained ResNet18, and global and local texture information of an expression sample is extracted; replacing a mask mechanism in the BERT model with a sorting mechanism; the LSTM is adopted to capture the time sequence change of the sample based on the DenseSift key point expression; fusing decision probability values based on global, local and time sequence information, and perfecting decision information extraction; and combining a cross entropy loss function and chi-square estimation to realize end-to-end model learning. According to the method, the robustness to face shielding, head posture change and uneven illumination is enhan