Face Expression Recognition Based On Convolutional Neural Network Combined With Attention Mechanism

Facial expression is one of the most powerful,the most natural and commonest signals for humans to express emotions and mental states.As a research hotspot in computer vision and human-computer interaction,facial expression recognition has attracted more and more attention in academia and industry in recent years.It is widely used in many fields,including medical monitoring,education evaluation,criminal investigation,emotional computing,and driving fatigue monitoring,etc.Although convolutional neural network has achieved good results in the field of facial expression recognition,the traditional convolutional neural network still has the problems of complex network models,large amount of model parameters and single feature information,which makes it can only be used in specific scenes.In addition,in the actual scene of facial expression recognition,the backgrounds of image are often complex,there are many useless feature information that can lead to low recognition rate of facial expression.In response to these problems,this thesis improves the facial expression recognition algorithm.The main research work includes the following points:(1)A face expression recognition algorithm based on Mobile Net V2 fused SENet is proposed.The algorithm uses a lightweight convolutional neural network model Mobile Net V2 and embeds a SENet to learn key facial features to improve the representational capability of the network.In addition,the linear bottleneck layer in the Mobile Net V2 model is improved to reduce the image features lost due to the increase in the number of network layers.(2)A face expression recognition algorithm based on multi-scale Xception fused CBAM is proposed.The algorithm adopts Xception network of lightweight convolutional neural network as the base model,and uses multi-scale depthwise separable convolution modules with convolutional kernels of 3×3,5×5,and 7×7 for feature extraction and fusion to extract richer facial expression feature information.In addition,the CBAM is embedded on the basis of multi-scale convolutional neural network,which makes the network model more focused on channel and spatially differentiated facial expression features.The improved algorithms proposed in this thesis is trained,tested and validated based on Fer2013 and KDEF face expression database.The experimental results show that the improved algorithm enhances the characterization ability of the network and improves the accuracy of face expression classification recognition,which verifies the effectiveness of the improved algorithm proposed in this thesis.