一种新的GBPA生成方法及其在模式识别中的应用

A novel method to determine GBPA and its application in pattern recognition

广义证据理论是一种在不完备识别框架中处理多传感器信息融合问题的实用方法.由于时代环境的影响,人们的认知存在局限性,难免会将不完备的识别框架认为是完备的,经典证据理论在这种情况下并不完全适用.因此,根据广义证据理论提出一种新的广义基本概率赋值(generalized basic probability assignment, GBPA)生成方法.该方法首先根据训练数据分别构造样本类别和测试样本的广义三角模糊数模型;然后通过计算样本和类别间的广义三角模糊距离生成GBPA;最后使用广义组合规则融合所有证据并得出最终的结论. Iris数据集的实验结果表明所提方法合理有效,即使在样本不足的情况下仍有较高的分类精度.

Generalized evidence theory(GET)is a useful method to address the problem of multi-sensor information fusion over the incomplete framework of discernment(FoD).Due to the limitations of cognition in the era,people inevitably considered the incomplete FoD as the complete FoD,and the classical evidence theory is not fully applicable in this case.Therefore,a new generalized basic probability assignment(GBPA)determination method based on the GET is proposed.According to the training data,the method first generates the generalized triangular fuzzy number(GTFN)models of the classes and the test samples,respectively.Then the GBPAs are determined by calculating the generalized triangular fuzzy distance between samples and classes.Finally,the generalized combination rule fuses all the bodies of evidence to obtain the final conclusion.The experimental results on the Iris dataset show that the proposed method is reasonably effective and has relatively high classification accuracy even in the case of insufficient samples.