摘要
采用人脸识别算法可以快速识别出教室中每个学生的身份,以掌握学生的出勤率信息,从而达到提升教学管理效率,促进教学质量提高的目的.为了克服人脸识别算法应用在课堂环境中识别准确率低的问题,本文提出了一种改进的InsightFace人脸识别算法.该算法基于MobileNet V2网络结构,将带有卷积注意力模块的CBAM-MobileNet网络结构代替InsightFace算法的ResNet人脸嵌入网络,同时采用限制对比度自适应直方图均衡化方法对待识别数据进行预处理,从而提高人脸图像的质量,提升了人脸识别准确率.改进后的算法在LFW和AgeDB-30数据集上测试准确率分别达到98.75%和88.60%,并且使用采集自课堂环境的Smart-Classroom数据集分别对教室中前后排学生的大、中、小三种尺寸人脸进行测试,分别较原算法准确率提高0.1%、2.6%、8.0%.
The face recognition algorithm can be used to quickly identify each student in the classroom in order to grasp the student attendance information, thus achieving the purpose of improving teaching management efficiency and promoting teaching quality improvement. In order to overcome the problem of low recognition accuracy of face recognition algorithm application in classroom, an improved InsightFace face recognition algorithm is proposed in this paper. The algorithm is based on the MobileNet V2 network structure, and the CBAM-MobileNet network structure with convolutional attention module replaces the ResNet face embedding network of the InsightFace algorithm, while the restricted contrast adaptive histogram equalization method is used to pre-process the data to be recognized, so as to improve the quality of face images and enhance the face recognition accuracy. The improved algorithm was tested on LFW and AgeDB-30 datasets with accuracy of 98.75% and 88.60%, respectively, and tested using Smart-Classroom dataset collected from classroom environment for large, medium and small size faces of students in front and back rows in the classroom, respectively, with accuracy improvement of 0.1%, 2.6% and 8.0% over the original algorithm.
作者
王宸
刘剑飞
郝禄国
曾文斌
曹丹
Wang Chen;Liu Jianfei;Hao Luguo;Zeng Wenbin;Cao Dan(School of Electronic Information Engineering,Hebei University of Technology,Tianjin 300401,China;School of Information Engineering,Guangdong University of Technology,Guangzhou 510006,China;R D department,Guangzhou Hison Computer Technology Co.Ltd.,Guangzhou 510663,China)
出处
《南开大学学报(自然科学版)》
CAS
CSCD
北大核心
2022年第2期59-68,共10页
Acta Scientiarum Naturalium Universitatis Nankaiensis
基金
河北省高等教育教学改革研究与实践(2018GJG060)
河北省研究生示范课程项目(KCJSX2020014)
广州科技计划项目(201802020008)
河北省高等学校科学技术研究重点项目(ZD2017021)。
关键词
卷积神经网络
深度学习
人脸识别
注意力机制
直方图均衡
convolutional neural network
deep learning
face recognition
attention mechanism
histogram equalization