基于密度特征与KNN算法的最优特征维数选择Optimal feature dimension selection based on density feature and KNN algorithm
孙国栋,梅术正,汤汉兵,周振
摘要(Abstract)：
为了保证基于同步触发双相机的仪表复杂字符识别中误识率为0,采用K最近邻算法对仪表字符特征进行训练分类,结合字符自身特点,提出最优特征提取与高宽维度选择方法,并设计实验获取1~4 096维密度特征的误识率与运行时间。实验结果表明,图像的密度特征总维度在230~260,高宽维度比接近1.4时,误识率为0的概率最大。该规律对采用KNN算法进行分类识别时最优密度特征维数选择具有一定指导意义。
关键词(KeyWords)： 复杂仪表;特征维数;误识率;KNN算法;密度特征;最优特征
基金项目(Foundation): 国家自然科学基金项目资助(51675166);国家自然科学基金项目资助(51205115)~~
作者(Author): 孙国栋,梅术正,汤汉兵,周振
DOI: 10.16652/j.issn.1004-373x.2018.16.020
参考文献(References)：
- [1]王达,杨朋威,张平,等.感应电条件下的电流互感器误差测试方法分析[J].内蒙古电力技术,2017,35(2):75-78.WANG Da,YANG Pengwei,ZHANG Ping,et al.Analysis of current transformer error test method under induced electricity[J].Inner Mongolia electric power,2017,35(2):75-78.
- [2]DODIG I,SRUK V,CAFUTA D.Reducing false rate packet recognition using dual counting bloom filter[J].Telecommunication systems,2017(1):1-12.
- [3]罗辉武,唐远炎,王翊,等.基于结构特征和灰度特征的车牌字符识别方法[J].计算机科学,2011,38(11):267-270.LUO Huiwu,TANG Yuanyan,WANG Yi,et al.License plate character recognition based on structural features and grayscale pixel features algorithm[J].Computer science,2011,38(11):267-270.
- [4]孙中悦,周天荟.基于特征点统计特征的人脸识别优化算法[J].北京交通大学学报,2015,39(2):128-134.SUN Zhongyue,ZHOU Tianhui.A face recognition optimization algorithm based on statistical characteristics of feature points[J].Journal of Beijing Jiaotong University,2015,39(2):128-134.
- [5]张爱娟.基于结构特征点的字符分割技术的研究[D].西安:西安电子科技大学,2014.ZHANG Aijuan.Research on character segmentation based on structure feature points[D].Xi’an:Xidian University,2014.
- [6]张彤,肖南峰.基于BP网络的数字识别方法[J].重庆理工大学学报(自然科学版),2010,24(3):43-47.ZHANG Tong,XIAO Nanfeng.Character identification based on a BP neural network[J].Journal of Chongqing University of Technology(Natural science),2010,24(3):43-47.
- [7]刘雨康,张正阳,陈琳琳,等.基于KNN算法的改进的一对多SVM多分类器[J].计算机工程与应用,2015,51(24):126-131.LIU Yukang,ZHANG Zhengyang,CHEN Linlin,et al.Improved KNN-based 1-vs-all SVM multi-class classifier[J].Computer engineering and applications,2015,51(24):126-131.
- [8]ZHANG S,LI X,ZONG M,et al.Learning k for KNN classification[J].ACM transactions on intelligent systems&technology,2017,8(3):1-19.
- [9]华辉有,陈启买,刘海,等.一种融合K-means和KNN的网络入侵检测算法[J].计算机科学,2016,43(3):158-162.HUA Huiyou,CHEN Qimai,LIU Hai,et al.Hybrid K-means with KNN for network intrusion detection algorithm[J].Computer science,2016,43(3):158-162.
- [10]KE Xizheng,LI Shiyan.Experimental study on optical axis alignment of the optical antenna under the spot broken[J].Acta Photonica Sinica,2017,46(4):1-9.