Research On Facial Expression Recognition Method Of Ward Patients

In recent years,the ageing population has become more prominent and the pressure on the lack of medical resources has been increasing,making the efficiency of patient care on wards a pressing issue.The facial expression recognition method based on deep learning has the problems of large number of participants,high computational resource consumption and low recognition accuracy.In view of the above problems,this paper conducts a research on the improvement of the ward patient facial expression recognition method based on deep learning,and the main research contents are as follows:(1)A patient expression recognition method based on Mobile Net V3 backbone network is studied for the problems of large number of parameters and large consumption of computational resources in ward patient expression recognition algorithm models.Firstly,the number of layers of Mobile Net V3 network is reduced by convolutional layers to effectively reduce the number of parameters;at the same time,the number of intermediate channels and output channels of the inverse residual structure are increased to 1.5-3.2 times of the original one to improve the feature information extraction capacity;finally,Mish is used instead of Hardswish activation function to realize the non-linearization after feature extraction.The experimental results show that the number of model parameters is reduced by 19.00% and the inference time is reduced by 11.93%after the improvement.(2)To address the issue of low recognition accuracy of patient expressions in Mobile Net V3 backbone network,an improved recognition method based on coordinated attention mechanism,CCA-Mobile Net,is investigated.Secondly,based on the characteristics of depth-separable convolution and inverse residual modules,the conditional coordinated convolution module CCA-bneck is designed and implanted into the backbone network.The experimental results show that the accuracy on the RAF-DB,FERPlus and My DB datasets are 85.51%,88.61% and95.38% respectively,with an accuracy improvement of 1.07%,0.60% and 0.97%respectively compared with that before the improvement.(3)To address the problem of low recognition rate of some patient expression categories due to severe imbalance of category samples in the RAF-DB and FERPlus datasets,the loss function is optimized and a patient expression recognition method based on fused multidimensional loss functions is investigated.Firstly,the sample category loss is designed,and the corresponding weights of the categories are divided according to the percentage of the sample categories;secondly,the central loss function is improved by adding channel domain and spatial domain attention mechanisms as central loss input values,increasing the inter-class distance while reducing the intra-class distance;finally,the category loss and central loss are fused by weight assignment,and the fusion results are incorporated into the CCA-Mobile Net model.The experimental results show that the accuracy of the less-sample category expressions of fear and disgust on the RAF-DB dataset is increased by 10% and 8% respectively,and the accuracy of the less-sample category expressions of contempt,disgust and fear on the FERPlus dataset is increased by 2%,2% and 4% respectively,which verifies the effectiveness of the proposed loss function.In summary,this paper proposes an improved lightweight ward patient facial expression recognition method,and validates the efficiency of the proposed method on public and homemade datasets,which can provide a practical and reliable option for ward patient facial expression recognition.