| One of the most important aspects in the field of artificial intelligence is the computer’s understanding of human emotions.Facial expressions are an important way for humans to express emotions and transmit emotions.With the development of science and technology,facial expression recognition technology can be applied to many fields such as smart security,smart medical care,digital entertainment,and smart campus.The core technology of facial expression recognition is the feature expression of images,and the purpose is to extract features with strong robustness and high recognition rate.The traditional facial expression recognition algorithm has complex process,poor performance and uncertainty.Fine-grained decoupled representation learning can capture small class gaps and large intra-class differences in images and assign them to corresponding representation data to achieve fine-grained image recognition.This paper focuses on the face analysis technology based on Fine-Grained decoupling representation,and conducts the following research:(1)Aiming at the problem that simple convolution is easy to cause information loss in the process of facial expression recognition using convolutional neural network,a fuzzy-optimized CNN-RNN facial expression recognition algorithm is proposed.Firstly,the input face image is scaled by the bilinear interpolation method,and the image sample data is expanded by the affine transformation technique.Then fuzzy logic is used to identify the features and deal with the nonlinear relationship between the features.Finally,the fuzzy control model is fed into a recurrent neural network(RNN)for loop iteration to classify different facial expression images.Tested and verified on international mainstream open datasets,the results show that the performance of the model is better than the existing CNN-Based methods,and the overall recognition rate is improved by up to 3.4%,indicating that the method is feasible and effective.(2)In order to further improve the network performance,a CNN-RNN facial expression recognition algorithm(ACNM)incorporating attention mechanism is proposed.Firstly,the channel attention module and the spatial attention module in the convolutional block attention mechanism module(CBAM)plus shortcut connection are improved into a lightweight convolutional residual attention module(CRAM).Then,the lightweight attention module CRAM is integrated into the fuzzy optimized CNNRNN network model extraction layer to improve the feature extraction performance and enhance the representation of expression discriminative features.Tested and verified on international mainstream open datasets,the results show that the new network model ACNM only increases the number of parameters by 0.28 M,and the recognition rate increases by 0.45%.The research shows that the problem of feature graph loss can be effectively avoided in feature extraction by fusing fuzzy control and optimized recurrent neural network in convolutional neural network with the premise of increasing the number of data sets,the application of global information is realized,and the generalization of network is enhanced,so the precision of network model detection is improved.The lightweight attention mechanism model can extract facial expression features more effectively without increasing the amount of model parameters too much.The experimental results show that the two facial expression recognition network models proposed in this paper are effective in improving the facial expression recognition rate. |
