基于多源测量与属性混合信息的分类识别方法  

Classification and Recognition Method of Information Based on Multi-source Measuringand Attributes Mixing

作　　者：关欣[1] 常进[1] 王虹[1] 衣晓[1] 

机构地区：[1]海军航空工程学院电子信息工程系,烟台264001

出　　处：《计算机与数字工程》2016年第4期615-620,共6页Computer & Digital Engineering

基　　金：国家自然科学基金重点项目(编号:61032001);教育部新世纪优秀人才支持计划项目(编号:NCET-11-0872)资助

摘　　要：充分利用多传感器系统提供的多源异类信息,将观测数据处于数据级的属性和处于特征级的属性混合作为描述目标的特征矢量;对特征矢量进行了主成分分析,在此基础上转换至三维直角坐标系寻找最优分类平面进行分类识别;采用"一对一"策略解决多类分类问题;通过仿真实验验证了该方法在加入不同百分比高斯白噪声环境下的有效性,并与BP神经网络识别方法在同等条件下作了对比实验,突出了论文所提方法正确识别率高、识别速度快和稳定性高的优越性。Multi-source and heterogeneous information provided by multisensor system is being utilized fully.The paper merges attributes whose observation data is at data level and somes at feature level into a feature vector that describes target.On the basis of principal component analysis of feature vector,it is transformed to triangular rectangular-coordinates system to find a optimal separating plane for classification and recognition.＂One-Against-One＂strategy is used to deal with the multi-class problems.The validity of the method is validated by simulation experiments in the environment of different percentages of gaussian white noise,then this paper carrys on comparison experiment with BP neural network recognition method in same condition.It shows the superiority of higher recognition rate,faster recognition speed and higher stability of the proposed method.

关 键 词：主成分分析 最近顶点规则 最优分类平面 BP神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]