Abstract：A new algorithm named Uncorrelated Discriminant Neighborhood Preserving Projections (UDNPP) is proposed based on manifold learning. And UDNPP algorithm includes the advantages of Linear Discriminant Analysis (LDA) and Neighborhood Preserving Projections (NPP). Actually, UDNPP attempts to preserve the geometry of neighborhoods, while maximizing the between-class distance. Moreover, the features extracted are statistically uncorrelated by introducing an uncorrelated constraint. Thus the interference from redundant information are reduced. The experimental results from millimeter wave radar target recognition show that UDNPP algorithm can give competitive results in comparison with current popular algorithms.
2010, Vol. 32 
