基于多特征和多阶段的雷达距离像识别

讨论了用新的识别系统识别高分辨雷达距离像 ( HRRP)。它分为两个阶段 :( 1 )在 K- L变换对距离像进行特征压缩后 ,用聚类法进行识别或采用某个规则排除某些不可能的种类。 ( 2 )对于第一阶段不能识别的模式 ,用种类少时效果较好的最佳鉴别准则对模式进行特征提取 ,并用 BP网络组作为分类器进行识别。实验表明该方法比常规的单分类器的方法有更高的识别率。

RADAR RANGE PROFILE IDENTIFICATION BASED ON MULTIPLE FEATURE AND MULTISTAGE CLASSIFIER

A range profile identification system based on multiple classification is discussed.Two processing stages are adopted, viz.:(1) after feature extraction with K L transform, a rough identification result or elimination of impossible candidates can be achieved by clustering as the first stage,(2) the remainder is identified by a BP network group with feature extraction of the optimal set based on Fisher criterion as the second stage.Experiments applying the data of a high resolution radar show that this mode of classication can be more effective than those using a single classifier.