Research On Facial Expression Recognition Method Based On Feature Fusion And Attention Mechanism

Facial expressions play a crucial role in interpersonal communication.At present,facial expression recognition has been widely used in human-computer interaction,teaching evaluation,traffic safety and other fields.However,the conventional convolutional neural network has certain limitations in facial expression recognition,including inadequate feature extraction and low accuracy in uncontrolled environments.Therefore,it is a very challenging task to design a network model that can extract more comprehensive and abundant feature information.To solve the above problems,a deep learning network is designed in this paper and named Feature Fusion and Attention Mechanism Network(FFAM-Net).FFAM-Net mainly consists of Res Net18＿Attention Feature Extraction Network,Local Feature Extraction Module,and Expression Uncertainty Module.The innovative points of the proposed method are as follows: 1)Since the Res Net18 Convolutional Neural Network cannot extract discriminative facial features,a new Res Net18＿Attention Feature Extraction Network is proposed in this paper.The network can extract the fusion features related to the facial expression region from both channel and space perspectives,to improve the overall performance of the network model.2)Facial expression recognition in the natural environment is easily affected by factors such as occlusion and posture change,so extracting only the global features of facial expression will have a certain impact on the effect of facial expression recognition.Therefore,this paper proposes a Local Feature Extraction Module,which can extract local features of facial expression images and make up for the deficiency of global features,to effectively solve problems such as occlusion and posture change in the field of expression recognition.3)In this paper,a new CF＿Loss function is designed,which consists of the Cross-Entropy function and the Focal Loss function.Based on the Cross-Entropy function,an adjustment factor is added to the Focal Loss function to reduce the weight of easily classified samples,so that the Expression Uncertainty Module can focus on indistinguishable samples,and finally improve the overall performance of the FFAM-Net network model.To verify the effectiveness of the FFAM-Net model proposed in this paper,this paper conducts related experiments on two public large-scale static facial expression image datasets RAF-DB and FERPLUS.The experimental results demonstrate that the proposed FFAM-Net outperforms the standard method and some other mainstream facial expression recognition methods,achieving superior results for facial expression recognition.