Research On Low-resolution Facial Expression Recognition Based On Lightweight Convolutional Neural Network

In human daily life,facial expressions play an important role in the transmission of non-verbal information.With the rapid development of artificial intelligence,the research of Facial Expression Recognition(FER)has achieved fruitful results in the fields of psychology,computer vision and pattern recognition.In addition,the automatic recognition of facial expressions plays an important role in the field of human-computer interaction.In the era of rapid development of artificial intelligence and Internet of Things technology,if robots can automatically recognize human facial expressions,it will greatly promote the development of human-computer interaction.Therefore,the realization of automatic recognition of facial expressions has important research significance.With the continuous advancement of computer vision technology,the application in the industry has become more and more extensive,the role of facial expression recognition is becoming more and more important,and the requirements for facial expression recognition are getting higher and higher.Especially in the mobile Internet era,in order to apply facial expression recognition to mobile devices or embedded devices,the size of the model is limited,and higher requirements are put forward for the real-time recognition and robustness of the model.In the natural environment,the face images collected by the camera are often of low resolution due to the influence of the collection equipment and other aspects.However,some current mainstream facial expression recognition solutions can only recognize high-resolution facial expressions,but in the case of low-resolution,the recognition effect is not ideal.In response to the above problems,this paper first proposes a two-channel facial expression recognition method based on a lightweight convolutional neural network.First,it detects and locates the facial expression image,crops out the non-face area,and performs data augmentation and preprocessing.Process,and then perform feature extraction on the grayscale image and LBP(Local Binary Patterns)image respectively,and then use feature dimensionality reduction and feature fusion,and we introduce the island loss function to solve the small difference between facial expression classes and the difference within the class Big problem.In addition,in order to recognize lowresolution facial expression images,this paper proposes a low-resolution facial expression recognition model based on coupling mapping,which uses a dual-branch structure and uses feature extraction subnets and super-resolution subnets.In the upper and lower branches,the high and low resolution facial expression images are processed respectively,and the two feature vectors obtained are projected into the same common feature subspace,and the gradient descent method and the error back propagation algorithm are used to update the convolutional neural network.The weight is used to ensure that the distance between high and low resolution facial expression images in the common feature subspace is minimized,and at the same time,it can be classified into the correct category.In order to verify the performance of the dual-channel facial expression recognition method based on lightweight convolutional neural network and the low-resolution facial expression recognition method based on coupling mapping,this paper conducts experiments based on JAFFE,FER2013,and CK+ databases.The experimental results show that the two-channel facial expression recognition method based on lightweight convolutional neural network designed in this paper achieves an accuracy of 95.238% on the JAFFE database.In the case of low resolution,the low-resolution facial expression recognition method based on coupling mapping designed in this paper has a good recognition effect on low-resolution facial expression images,especially at the ultra-low resolution of 6×6,which improves the effect obvious.