Research On Face Expression Recognition Based On Deep Learning Low Resolution

Facial expression recognition has been a popular research topic in the field of computer vision as it is a common way to convey emotions and information through facial expressions in people’s daily communication.In recent years,with the continuous development of artificial intelligence,computer vision has made breakthroughs in the field of image recognition,and research on facial expression recognition is gradually gaining ground,with promising applications in clinical medicine,safe driving,intelligent education and fraud analysis.Face expression data sets are easily affected by factors such as lighting,background and pose when collected,resulting in models with insufficient ability to represent expression features and low recognition accuracy;and with the simple stacking of neural network layers,when there are too many layers,a gradient explosion will occur,resulting in an unstable model that cannot be trained effectively;when the amount of parameters of the network model is too large,it is not possible to effectively When the number of parameters of the network model is too large,the model cannot be effectively applied to edge-end devices,and there is also a risk of overfitting the model.In order to solve the above problems,the main work of this thesis includes two aspects:1.To address the problem of insufficient representation capability of the local perceptual field of facial expressions,this thesis designs a coordinate attention mechanism module to redistribute the weights of feature maps in both horizontal and vertical dimensions to extract deeper local information of facial expressions,enriching the model’s representation capability of facial expression features and improving the accuracy of model recognition,while optimising the pyramidal convolution structure to extract multi-scale The method also optimizes the pyramid convolution structure to extract the multi-scale feature information of face expressions,which expands the perceptual field of the model,and experiments prove that the method effectively improves the accuracy of the model.2.For the problem of small face expression dataset,this thesis performs image preprocessing operations on the original dataset,and through improved data enhancement methods,such as random erasure,rotation,flip and vertical mirroring,the overall number of databases is effectively expanded to reduce the impact of factors such as small face expression dataset on the model,and the robustness of the model is enhanced.In order to reduce the number of parameters in the network and reduce the redundancy of the model,we also accelerate the speed of the network and improve the non-linear expression capability of the model.In order to verify the effectiveness of the proposed algorithm,experiments were conducted on public datasets such as FER2013,CK+ and JAFFE respectively.Compared with the state-of-the-art algorithms,the proposed algorithm achieves better recognition and higher accuracy while maintaining a faster computational speed.