结合SVM和DS 证据理论的多极化HRRP 分类研究

针对雷达目标一维距离像(HRRP)识别问题,结合支持向量机(SVM)和 DS 证据理论提出一种多极化HRRP 分类方法—–SDHRRP.该方法通过混淆矩阵获取基分类器之间的距离,从而根据基分类器对不同目标类的分类能力给其赋予不同的可信度.将该可信度值与 SVM 后验概率结合到 DS 证据理论的基本概率赋值(BPA)中,以实现 SVM 和 DS 证据理论在目标识别中的有效结合.对实测目标数据的实验结果表明,基于分类器可信度得到的 BPA 能够有效避免证据冲突, SDHRRP 方法可以有效降低融合分类的误差率.

Multi-polarized HRRP classification by SVM and DS evidence theory

For the hotspot of high resolution range profile(HRRP) usage in radar target recognition, a multi-polarized HRRP
classification approach combing SVM and DS evidence theory‘SDHRRP’is presented. The method defines different
confidence for the classifiers based on the distance between each other given by confusion matrix. Then the value and the
posterior probability of SVM are integrated into the BPA(basic probability assignment), which achieves the combination of
SVM and the evidence theory in target recognition. The gained BPA based on classifier reliability can avoid the evidence
conflict efficiently. The experimental results based on the measured data show the effectiveness of the proposed approach.