长春工程学院学报（自然科学版）2024，Vol.25Issue(1)：59-63,5.DOI:10.3969/j.issn.1009-8984.2024.01.011
基于注意力多尺度融合的人脸表情识别算法研究
Research on Facial Expression Recognition Algorithm Based on Attention Multi-Scale Fusion
摘要
Abstract
The application of information technology in teaching leads to a lack of emotional communication between teachers and students.In order to compensate for the lack of emotional communication during the teaching process and obtain better teaching feedback,a facial expression recognition algorithm based on at-tention mechanism and multi-scale feature fusion(ASMF)is proposed.The algorithm uses Resnet 50 as the backbone network.It firstly fuses the output characteristics of multi-layer convolutional neural net-works at multiple scales,introduces contextual information while extracting richer and more effective ex-pression feature information.Secondly,the attention mechanism is integrated into the network,and through weighted learning of each channel,attention feature maps are obtained to enhance the expression a-bility of features and suppress the impact of redundant information.Then,the Dropout mechanism and Softmax Loss function are added to further improve the discriminability of the extracted facial features Fi-nally,the effectiveness and stability of the algorithm are validated by using ablation experiments on both publicly available datasets and self-made student classroom expression datasets,with a recognition accuracy of 93.87%.关键词
表情识别/深度残差网络/注意力机制/多尺度融合Key words
facial expression recognition/deep residual network/attention mechanism/multi-scale fusion分类
信息技术与安全科学引用本文复制引用
安毅,张慧,陈思秀,郑文..基于注意力多尺度融合的人脸表情识别算法研究[J].长春工程学院学报（自然科学版）,2024,25(1):59-63,5.基金项目
吉林省职业教育科研课题项目(2023XHY262,2023XHZ016)吉林省科技发展计划项目(20220203178SF)吉林省高等教育教学改革研究课题(2024L5LY26U0058) （2023XHY262,2023XHZ016）
