摘 要：为了及时、客观、准确地在课堂上了解每个学生的听课状态，采用一种全局多尺度和局部注意力网络（MA-Net） 的表情识别模型，具体来说，模型由三个部分组成：图像预处理及特征提取、全局多尺度模块和局部注意力模块。图像预处理提高输入图像的质量要求，核主成分分析进行特征提取，全局多尺度模块融合不同感受野的特征，降低深度卷积对遮挡和非正面姿态的敏感性，而局部注意力模块可以引导网络专注于局部显著特征，同时定义一个 circle 损失函数以规范整个学习过程，模型在FED-RO 测试集上得到了较高的准确率。
关键词：听课状态；多尺度；注意力网络；circle 损失函数
DOI:10.19850/j.cnki.2096-4706.2022.18.017
课题项目：长春市教育科学“十四五”规划 2021 年度一般课题项目（JKBLX2021071）
中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2022）18-0071-05
Research on Students’ Emotional State Based on Expression Recognition Technology—Taking the Junior Middle School Classroom as an Example
JIAO Shuang1, YAN Yuxing2
(1.Changchun Institute of Education, Changchun 130033, China; 2.Changchun University of Science and Technology, Changchun 130022, China)
Abstract: In order to timely, objectively and accurately understand the listening status of each student in the classroom, a global multi-scale and local attention network (MA-Net) expression recognition model is adopted. Specifically, the model consists of three parts: image preprocessing and feature extraction, global multi-scale module and local attention module. The image preprocessing improves the quality requirements of input images, the kernel principal component analysis performs feature extraction, the global multi-scale module fuses features from different receptive fields, and reduces the sensitivity of deep convolution to the occlusion and non-frontal poses. The local attention module can guide the network to focus on local salient features, at the same time, it defines a circle loss function to standardize the entire learning process, and the model obtains a higher accuracy rate on the FED-RO test set.
Keywords: listening status; multi-scale; attention network; circle loss function
参考文献：
[1] FERNANDEZ P D M，PEÑA F A G，REN T I，et al. FERAtt：Facial Expression Recognition with Attention Net [C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops（CVPRW）.Long Beach：IEEE，2019.
[2] KUMAWAT S，VERMA M，RAMAN S. LBVCNN：Local Binary Volume Convolutional Neural Network for Facial Expression Recognition from Image Sequences [C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops （CVPRW）. Long Beach：IEEE，2019：207-216.
[3] LEE J，KIM S，KIM S，et al. Multi-Modal Recurrent Attention Networks for Facial Expression Recognition [J].IEEE Transactions on Image Processing，2020，1（29）：6977-6991.
[4] LI R，TIAN J，CHUA M.C.H. Facial expression classification using salient pattern driven integrated geometric and textual features [J]. Multimed Tools and Applications，2019，78（20）：28971-28983.
[5] PUN T. A new method for grey-level picture thresholding using the entropy of the histogram [J]. Computer Vision Graphics & Image Processing，1980，2（3）：223-237.
[6] SCHÖLKOPF B，SMOLA A J，Müller K R. Kernel Principal Component Analysis [C]//Artificial Neural Networks - ICANN’97， 7th International Conference，Lausanne：Springer Verlag，1997： 555-559.
[7] 王军，高智勇，刘海华，等 . 基于 KPCA 和 Gabor 小波的特征融合人脸识别 [J]. 现代科学仪器，2010（3）：9-13.
[8] WOO S，PARK J，LEE J Y，et al. CBAM：Convolutional Block Attention Module [J].Springer，2018：3-19.
[9] LI Y，ZENG J B，SHAN S G，et al. Patch-Gated CNN for Occlusion-aware Facial Expression Recognition [C]//2018 24th International Conference on Pattern Recognition（ICPR）.Beijing： IEEE，2018：2209-2214.
[10] LI Y，ZENG J B，SHAN S G，et al. Occlusion Aware Facial Expression Recognition Using CNN With Attention Mechanism [J].IEEE Transactions on Image Processing，2019，28（5）：2439-2450.
作者简介：焦爽（1991—），女，汉族，吉林公主岭人，助教，硕士，研究方向：图像处理；闫禹行（1990—），男，满族，吉林四平人，初级工程师，硕士，研究方向：图像处理。