基于证据K近邻的目标识别新方法

A New Target Recognition Method Based on Evidence Theoretic K-NN Rule

摘要: 本文提出了一种基于证据K近邻的目标识别新方法，在Zouhal改进KNN算法的基础上增加了训练修正步骤。首先，求得每一个目标类别的参考最近邻距离，使训练样本中该目标类别的样本在经验风险最小化的前提下与其他样本完成分离；然后，利用求得的参考最近邻距离和证据理论结合得出初始的识别分类结果；第三，设置混淆矩阵P，通过神经网络寻优迭代，获得P矩阵参数，用于Zouhal分类结果修正；最后，通过多数据集验证了P矩阵的泛化能力，通过与经典算法的分类精度对比验证了新方法的可行性和有效性。

Abstract: This paper presents a new target recognition method based on evidence theoretic K-NN rule. A training correction step was added on the basis of Zouhal’s improved KNN algorithm. Firstly, a reference nearest neighbor distance of every target class should be computed in order to separate samples of one class from other samples with least error rate. Secondly, the initial classification result can be got through the reference nearest neighbor distance and DS evidence theory. Thirdly, setting up confusion matrix P and optimizing iteration through neural network to obtain matrix parameters for Zouhal’s classification result correction. Finally, the generalization ability of the matrix P is verified through multiple data sets, and the feasibility and effectiveness of the new method are verified by comparing with the classification accuracy of the classical algorithm.