摘要
本文深入研究了卷积神经网络和逻辑回归分类器,利用两者的优点提出了一种混合系统,将卷积神经网络和Logistic回归分类器进行串联,该模型进行了两步学习,首先训练卷积神经网络识别面部图像,接着用逻辑回归分类器对上一步训练的特征进行二次分类,同时将使用卷积神经网络的特征提取应用于规范化数据。最后人脸数据库中进行训练和测试,通过实验证明本文算法可以在更短的时间内提高分类率,在不同的光照下也能准确识别人脸表情,具有较好的稳定性和鲁棒性。
In this paper,the convolution neural network and the logic regression classifier are studied deeply,and a hybrid system is proposed by using the advantages of both.The convolution neural network and the logistic regression classifier are connected in series.The model is studied in two steps.First,the convolution neural network is trained to recognize the face image,then the logic regression classifier is used to classify the features trained in the previous step At the same time,the feature extraction using convolutional neural network is applied to the normalized data.Finally,the face database is trained and tested.The experimental results show that the algorithm can improve the classification rate in a shorter time,and can accurately recognize the face under different light conditions,with good stability and robustness.
作者
王偏
陈恳
WANG Pian;CHEN Ken(Faculty of Electrical Engineering and Computer Science,Ningbo University,Ningbo 315211,China)
出处
《无线通信技术》
2020年第1期38-42,共5页
Wireless Communication Technology
关键词
卷积神经网络
Logistic回归分类器
机器学习算法
反向传播和人脸识别
convolutional neural network
logistic regression classifier(LRC)
machine learning algorithm
back propagation and facial recognition
作者简介
王偏,女,1992年生,硕士,研究方向为人脸识别,宁波大学信息科学与工程学院。
