流形学习的提出和发展为毫米波雷达目标识别提供了新的思路。针对传统特征提取算法的不足, 提出了一种基于核的非线性流形学习算法,即核不相关邻域保持投影(KUNPP)。该算法在邻域保持投影的基础上引入再生核, 将数据映射到Hilbert空间; 在Hilbert空间内执行邻域保持投影算法, 并引入不相关约束, 使得到的特征向量具有不相关性, 减少冗余信息。将KUNPP应用于毫米波雷达目标识别, 仿真数据集和实测数据集的实验结果均表明算法能取得较好的结果。

A novel personal recognition system utilizing palm vein patterns and a novel technique to analyze these vein patterns is presented. The technique utilizes the curvelet transform to extract features from vein patterns to facilitate recognition. This technique provides optimally sparse representations of objects along the edges. Principal component analysis (PCA) is applied on curvelet-decomposed images for dimensionality reduction. A simple distance-based classifier, such as the nearest-neighbor (NN) classifier, is employed. The experiments are performed using our palm vein database. Experimental results show that the algorithm reaches a recognition accuracy of 99.6% on the database of 500 distinct subjects.