基于注意力机制的表情识别改进方法  

An Improved Method for Facial Expression Recognition Based on Attention Mechanism

作　　者：徐梦阳 XU Mengyang(Xi'an Technological University,Xi'an 710021,China)

机构地区：[1]西安工业大学,陕西西安710021

出　　处：《现代信息科技》2024年第8期102-105,110,共5页Modern Information Technology

摘　　要：针对人脸表情识别中存在的受到光照和姿势的影响导致识别精度不高和深度学习模型参数量巨大的问题,文章提出一种基于注意力机制的卷积神经网络改进模型。通过引入注意力机制模块,使模型选择性地关注目标对象的局部重要信息,降低无关信息的干扰;同时,利用较少神经元数量与大卷积核的神经网络,大幅减少了网络参数,该方法构建了一种层次更浅、参数量更少的轻量卷积神经网络模型。在CK+人脸表情数据集上进行实验,结果表明,文章提出的方法在保证人脸识别精度的情况下,还大大减少了模型参数,其准确率可达到96.37%。Targeting the problems of poor recognition accuracy and a large number of Deep Learning model parameters due to light and posture influence in facial expression recognition,this paper proposes an improved Convolutional Neural Network model based on Attention Mechanism.Through the introduction of the Attention Mechanism module,the model selectively focuses on the locally important information of the target object and reduces the interference of irrelevant information,while using a neural network with fewer neurons and a large convolutional kernel,the parameters of the network are significantly decreased,and the method builds a lightweight Convolutional Neural Network model with a shallower hierarchy and fewer parameters.Experiments are conducted on the CK+facial expression dataset,and results show that the proposed method significantly reduces model parameters while ensuring facial recognition accuracy,with an accuracy rate of 96.37%.

关 键 词：人脸表情识别 注意力机制 卷积神经网络 深度学习 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]