基于核不相关邻域保持投影的毫米波雷达目标识别

Millimeter Wave Radar Target Recognition Based on Kernel Uncorrelated Neighborhood Preserving Projects

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。

The proposal and development of manifold learning offer new thinking of millimeter wave radar target recognition. A nonlinear manifold learning algorithm using kernel technology, kernel uneorrelated neighborhood preserving projects （KUNPP）, is presented according to the deficiency of traditional feature extraction method. The algorithm maps the data into a Hilbert space by introducing reproducing kernel based on neighborhood preserving projects （NPP）. Then NPP is performed in the Hilbert space. And an uncorrelated constraint is introduced, which makes the feature vectors uncorrelated and reduces the redundancy of information. Millimeter wave radar target recognition experiments were performed using KUNPP, and the results from simulated data and measured data show that KUNPP can obtain better performance.