结合多分类RVM和DS的弹道目标HRRP融合识别方法

Ballistic Target HRRP Fusion Recognition Combining Multi-class Relevance Vector Machine and DS

摘要: 多特征融合可以有效提高目标识别正确率.本文将相关向量机（relevance vector machine，RVM）二类分类模型扩展为多类分类概率模型MRVM（multi-class relevance vector machine，MRVM），然后与DS证据理论相结合，用于弹道中段目标高分辨距离像（high resolution range profile，HRRP）分类和融合识别，提出了结合MRVM和DS的HRRP融合识别方法.该方法充分利用了多类分类RVM输出的概率信息，解决了用DS证据理论进行融合时基本概率赋值获取问题.仿真实验结果表明MRVM估计的样本后验概率更准确，融合识别后的正确率更高.

Abstract: Multifeature fusion can significantly improve the recognition performance of target discrimination. A multiclass RVM model, based on the basic RVM model, is extended and DS evidence theory is used to fuse the recognition result. A HRRP (high-resolution range profile) classification approach combining MRVM and DS evidence theory is then presented. The posterior probability of multiclass RVM is integrated into the BPA (basic probability assignment). This applies the combination of RVM and DS evidence theory to target recognition and solves the difficulty of getting BPA in DS evidence theory. Experiment results based on simulated data show tthat he probability estimated by MRVM is more precise and the fusion recognition performance is better, which testifies the efficiency of the proposed method.