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 per ‐centages of gaussian white noise ,then this paper carrys on comparison experiment with BP neural network recognition meth ‐od in same condition .It shows the superiority of higher recognition rate ,faster recognition speed and higher stability of the proposed method .%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。
展开▼
