Research On Facial Emotion Recognition Based On Deep Learning

As we all know,the expression of facial emotion occupies a very large part in human communication,and is also a means of human communication.With the rapid growth of artificial intelligence,various intelligent products have been applied to life,life intelligence has become a new label in the 21 st century,and the application of facial emotion recognition is gradually integrated into products and life,so that the major intelligent products are more intelligent,we can see the importance of facial emotion recognition in it.And although the traditional method for the study of facial emotion recognition has a certain effect,but the extracted features is mainly based on manual and error is bigger,and artificial intelligence in big background,the deep learning method for visual processing has entered a rapid development period,the neural network can be used to extract the features of a more comprehensive.Therefore,this paper conducts in-depth research on facial emotion recognition based on deep learning,and its specific work content is as follows:A face emotion recognition algorithm based on attention mechanism is proposed.With the training process of the network model,the network may produce "amnesia" for the original information due to the long time training.Therefore,this paper designs a face emotion recognition algorithm based on the attention mechanism,which mainly combines the attention module to focus on the key parts of the face.First of all,this paper uses Open CV to perform preprocessing operations of data enhancement on the data set,and the overall network model structure mainly includes three parts,namely the spatial transformer,the backbone network and the attention module.In the spatial transformer network part,this paper compares two transformer networks based on convolutional layers and fully connected layers,and finally,the transformer network based on convolutional layers is adopted;In the backbone network,Alexnet network is adopted as the benchmark network to replace 3*3convolution with 3*3,1*3 convolution kernel,3*1 convolution and other asymmetric convolution to enhance the backbone;Finally,improved attention modules are added in the first and last layers of the backbone network to focus on key parts of the face.Combining the above,a clear effect is obtained in the dataset.A facial emotion recognition algorithm with two-channel characteristics based on Involution operator is proposed.Because in the convolution,the improvement of the network model using the convolution kernel with space invariance and channel specificity is now in the white-hot stage,and this paper does the opposite using the channel invariance and space specificity of the involution operator and design two-channel two-feature convolutional network based on revolution operator and convolution operator.First of all,this paper designs the involution branch network using the involution operator by designing Alexnet network as the benchmark to replace its convolution kernel with the involution operator.Because of its channel invariance characteristics,1*1 convolution is added before each involution to change the channel,then based on this channel network and the convolutional neural network are connected in parallel to form a two-channel convolutional network,the obtained network structure has obvious effect on multiple datasets.Design and implement a facial emotion recognition system.Finally,this paper builds a facial emotion recognition system based on deep learning by using the Py Qt module,python language and the Py Torch framework,this system can realize the functions of model selection,video recognition and static image recognition,bringing users a multifunctional experience.