摘要
为了研究人眼状态识别对人眼定位的依赖性和实际使用中分类模型泛化能力不佳的问题,本文提出了一种基于选择性区域的卷积神经网络的融合决策的眼部状态识别方法。该方法用SR方法预处理wild人脸数据集,扩大了训练集的规模并引入了对眼部的先验知识,在此基础上训练卷积神经网络的分类模型进行眼部状态识别的评估。对比实验结果可知,基于SR-CNN融合决策的眼部状态识别方法测试的准确度能达到95%左右,显著降低了测试错误率,提高了模型的泛化能力和准确性。
To study the problem of eye state recognition,which includes dependence on eye location and poor generalization in practical application,this paper proposes an eye state recognition that uses a fusion-decision method based on convolutional neural networks(CNN) in selective region(SR). Firstly,the SR method was used to preprocess the wild face dataset to achieve data augmentation and thereby introduce prior knowledge about eyes. Then a CNN model was trained to evaluate error rate of the eye state recognition. The results of comparative experiment show that the accuracy of the eye state recognition method based on SR-CNN fusion decision method can reach about 95%,which significantly reduces the error rate and improves the generalization ability and accuracy of the model.
作者
黄斌
陈仁文
周秦邦
唐杰
HUANG Bin;CHEN Renwen;ZHOU Qinbang;TANG Jie(State Key Laboratory of Mechanics and Control of Mechanical Structures,Nanjing University of Aeronautics and Astronautics,Nanjing 210016,China)
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2018年第7期1233-1238,共6页
Journal of Harbin Engineering University
基金
国家自然科学基金项目(51675265)
江苏高校优势学科建设工程项目(PAPD)
关键词
SR方法
融合决策
卷积神经网络
眼部状态识别
过拟合
网络优化
数据增强
SR method
fusion decision
convolutional neural networks
eye state recognition
overfitting
networks optimization
data augmentation