应用于课堂场景人脸验证的卷积网络方法研究  

Research on convolutional network method applied to face verification in classroom scenes

作　　者：吴江林 刘堂友[1] 刘钊 Wu Jianglin;Liu Tangyou;Liu Zhao(School of Information Science and Technology,Donghua University,Shanghai 201620,China)

机构地区：[1]东华大学信息科学与技术学院,上海201620

出　　处：《信息技术与网络安全》2019年第5期42-47,共6页Information Technology and Network Security

摘　　要：对于人脸验证应用于课堂场景的问题,通过教室内的摄像头采集学生图像数据集,然而受光照、姿势和环境因素的影响,采集到的图像质量较低,一般的深度学习模型学习难度很大。针对这些问题,对采集到的图像进行了图像预处理,建立卷积图像分类模型与残差网络图像分类模型,并且修改损失函数,提高学习复杂度,训练出紧凑的人脸特征表达。设置了人脸验证阈值,实现人脸验证。通过实验分析在不同数据集上两个模型的精度,并验证修改的损失函数可改善模型性能,最后结果表明在采集到的图像数据集上正确率最高可以达到99. 97%,通过理论分析和实验证实了设计方法的有效性。For the problem of face verification applied to the classroom scenes,this paper collects the image dataset of students through the camera in the classroom.However,due to the influence of illumination,posture and environmental factors,the quality of collected image is low,the general deep learning model is very difficult to learn feature.Aiming at these problems,this paper preprocesses the acquired images,establishes a convolutional image classification model and a residual network image classification model,and modifies the loss function to improve the complexity of learning and train a compact facial feature expression.This paper sets the face verification threshold to implement face verification.Analyzing the results of the two models on different datasets and the effects of modified loss function through experiments. The final result can reach up to 99.97% accuracy on the collected image dataset.The validity of the designed method is verified by theoretical analysis and experiments.

关 键 词：课堂场景 人脸验证 深度学习 损失函数 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]