摘要
人脸口罩穿戴识别技术可以有效监督及管控人们佩戴口罩.本文基于迁移学习理论,共享经典卷积神经网络部分参数,修改其最后几层连接层,使用8967张图像样本集进行训练,得到了新模型;同时结合了人脸检测技术,针对检测后人脸子图像,采用图像分类方法实现了快速识别.通过迁移学习对深度网络模型开展迁移训练,解决了因为样本量少导致的准确率低等问题,新模型能够有效解决人脸口罩穿戴识别问题,使源领域知识得到了迁移.通过MATLAB编写迁移学习程序和应用仿真主程序,调用了摄像头硬件实现了真实场景应用仿真.实践证明,该研究具有较强的应用价值.
Face mask wearing recognition technology can effectively monitor and control people to wear the mask.Based on the transfer learning theory,the new model was obtained by sharing some parameters of the classical convolutional neural network,modifying its last several connection layers,training with 8967 image samples.At the same time,image classification method was used to realize the fast recognition of the face image of later generations combined with face detection technology.Migration training of deep network model was carried out through transfer learning,which solved the problem of low accuracy due to the small sample size.The new model could effectively solve the problems in epidemic prevention and control,and transfer the knowledge of source domain.The transfer learning program and the main application simulation program were wrote through MATLAB,and the camera hardware was called to realize the real scene application simulation.The research has proved that it has strong application value.
作者
刘微
宁维奇
LIU Wei;NING Wei-qi(College of Information and Technology,Jilin Normal University,Siping 136000,China;College of Information and Engineering,Shenyang Ligong University,Shenyang 110159,China)
出处
《吉林师范大学学报(自然科学版)》
2023年第1期96-103,共8页
Journal of Jilin Normal University:Natural Science Edition
基金
吉林省科技发展计划项目(20220101041JC)。
关键词
迁移学习
卷积神经网络
口罩穿戴识别
人脸检测
transfer learning
convolutional neural network
mask wearing recognition
face detection