多分类器融合实现机型识别

Combining Classifiers for Aircraft Type Recognition

针对空战目标识别中机型识别这一问题,提出了基于多分类器融合的识别方法。该方法以战术性能参数为输入,便于满足空战的实时性要求。通过广泛收集数据,得到机型识别的分类特征,选取分类特征的子集作为单分类器的特征,用BP网络设计单分类器,然后选用性能优良的和规则进行分类器融合,求得最终的决策。实验结果表明,多分类器融合的识别性能明显优于参与融合的分类器,也优于相同输入的单分类器。该方法的另一特点是能够进行缺省推理,因而有较强的抗干扰能力,适合真实战场环境的需要。

Aiming at aircraft type recognition in the field of automatic target recognition,a method based on combining classifiers is proposed.The proposed method is promising in meeting the rigid requirement for time in air warfare,for it adopts some tactical parameters as its input rather than graphics or images that are adopted in former research.Classification characteristics of aircrafts are obtained by searching materials widely.Authors design five BP network classifiers that use different subsets of characteristics as their input vectors.Decisions of these classifiers are combined by the“sum”rule to obtain the ultimate results.An experimental comparison of various classification schemes demonstrates that the recognition capability is improved by combining classifiers.The recognition method of combining classifiers outperforms each component classifier that participates combination and outperforms the single classifier that has the same input.Another characteristic of the proposed method is that it can perform default reasoning,which indicates that it can derive encouraging results even if some parameters are unavailable,while this case is usual in the actual battlefield.