Expression recognition method based on convolutional feature and Bayesian classifier

Aiming at the problem of prediction uncertainty in facial expression recognition by Convolutional Neural Networks (CNNs) due to data noise and sub - optimal hyperparameter settings, this paper proposes a hybrid model (VGG - BN+Bayes) that combines CNN feature extraction and Bayesian classifier to improve recognition accuracy and quantify uncertainty. Firstly, the folding layer of the CNN is used to extract deep - level features of facial images, and the scanning layer is combined to reduce the redundancy of model parameters. To address the output fluctuation problem of the CNN, a Bayesian classifier is introduced to replace the traditional Softmax layer. The reliability of weights is optimized through probability distribution estimation and multiple samplings. The CK+ dataset is used in the experiment. Strategies such as batch normalization and weight decay are adopted to prevent overfitting, and the learning rate is adjusted to optimize the convergence of the loss function. The results show that the accuracy of the improved model reaches 98.08% on the CK+ dataset. Comparative experiments verify that the embedding of the Bayesian classifier effectively reduces the prediction variance and enhances the robustness of the model in complex scenarios.