ODNPP降维算法在毫米波探测器目标识别中的应用

摘要

On the basis of neighborhood preserving projection algorithm, a new dimensionality reduction algorithm based on manifold learning named orthogonal discriminant neighborhood preserving projections (ODNPP) is proposed in this paper. The algorithm making full use of the class information of samples increases class scatter constraint in the objective function and intro-duces orthogonalization processing to calculate orthogonal projection matrix. ODNPP algorithm, NPP, NPDP and ONPP are ap-plied respectively in the simulation experiment of millimeter wave detector target recognition, and the experimental results show that ODNPP algorithm can find low dimensional manifold that embedded in the observed data of high dimensional space. ODNPP algorithm is used to reduce the dimensionality of the feature, and the reduced features can obtain higher recognition rates.%在邻域保持投影算法的基础上,研究了一种基于流行学习的维数约简算法,即正交判别邻域保持投影(ODNPP).该算法充分利用样本的类别信息,在目标函数中增加了类间散布约束,同时引入正交化处理,求取正交投影矩阵.将ODNPP算法与NPP,NPDP和ONPP三种算法分别应用于毫米波探测器目标识别仿真实验,结果表明ODNPP算法能够发现嵌入在高维观测数据空间中的低维流形,利用ODNPP算法降维后的特征可获得更高的识别率.