雷达高分辨距离像是目标的重要结构特征,其维数通常很高,造成数据可分性表达差,识别过程计算复杂度高,识别 率低。为降低距离像的维数,提出一种新的距离像特征提取方法,即采用直接线性判别分析(dLDA)在距离像幅度谱差分 空间进行特征提取,得dLDA幅度谱差分子空间。目标识别即在所得dLDA幅度谱差分子空间中进行。采用外场实测数 据,分别训练了最小距离分类器和one-against-all 支撑向量机分类器,2种分类器的识别结果均表明,该方法可显著地降低 数据维数并提高识别率。

Radar high resolution range profile (HRRP) is an important type of target structure feature. The dimensionality of HRRP is very high, which results in poor data discriminant representation, high recognition computation complexity, and low recognition rate. In order to reduce the dimensionality of HRRP, a novel HRRP feature extraction method is presented, i. e. , direct linear discriminant analysis(dLDA) is used to perform feature extraction in the amplitude spectrum difference space of HRRP, and a dLDA amplitude spectrum difference subspace is obtained. Target recognition is implemented in the resulting dLDA amplitude spectrum difference subspace. Shortest distance classifier and one-against-all support vector machine (SVM) classifier are designed to evaluate the recognition performance. Experimental results for measured data show that the presented method reduces data dimensionality greatly and improves recognition rate remarkably.

国家自然科学基金;中央高校基本科研 业务费专项资金;陕西省教育厅自 然科学专项基金;西安邮电学院博士启 动基金资助课题