基于熵的自适应加权投票HRRP融合识别方法  

HRRP fusion recognition by a self-adaptive weighted majority vote strategy based on entropy

作　　者：王晓丹[1] 李睿[1] 薛爱军[1] 孙向芳[2] 

机构地区：[1]空军工程大学防空反导学院 [2]中国人民解放军69026部队

出　　处：《系统工程与电子技术》2017年第4期707-713,共7页Systems Engineering and Electronics

基　　金：国家自然科学基金(60975026;61503407)资助课题

摘　　要：基于不同分类器对同一样本分类能力不同,同一分类器对不同样本可分程度不同的思想,为不同样本赋予不同融合权重,提出了一种基于熵的自适应加权投票高分辨距离像(high range resolution profile,HRRP)融合识别方法。该方法将二分类相关向量机(relevance vector machine,RVM)扩展为多类分类RVM概率模型,并对不同HRRP特征样本进行分类,利用每个多类分类RVM输出的样本后验概率信息计算出的熵值自适应为各个样本赋予权重,使得不同分类器以及同一分类器对不同样本的决策占有不同的比重,熵值越大的样本赋予的融合权重越低,最后通过加权投票方法实现融合识别,得到目标的最终识别结果。仿真实验结果验证了所提方法的有效性。Based on the novel idea of assigning different according to the classification ability and assigning different according to the separable degree of samples respectively, a on entropy for high range resolution profile （HRRP） fusion weights for the same sample by different classifiers weights for different samples by the same classifier selbadaptive weighted majority vote strategy based recognition is proposed. The multi-class relevance vector machine （MRVM） probabilities model is extended based on the basic RVM model, and three different MRVMs is used to classify different HRRP feature samples, then entropy calculated by the posterior probability of different MRVMs is used to assign weight adaptively, so that different classifiers and the same classifier occu- py different proportions in decision-making for different samples, low weight is assigned to sample with high en- tropy. Finally, the weighted majority vote strategy is used to fusion different feature classified results and get the final target recognition results. Experiment results based on simulated data show the efficiency of the pro- posed method.

关 键 词：目标识别 高分辨距离像 相关向量机 加权投票 熵 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]