基于BP神经网络和改进D-S证据理论的目标识别方法

TARGET RECOGNITION METHOD BASED ON BP NEURAL NETWORKS AND IMPROVED D-S EVIDENCE THEORY

针对应用经典D-S证据理论时,其关键参数基本概率赋值(BPA)往往凭主观经验获得,导致决策可信度低的问题,提出通过构建BP神经网络来获取基本概率赋值的方法。该方法利用BP神经网络强大的自学习和非线性映射能力,归一化输出值得到基本概率赋值。同时,为了解决高冲突度证据合成结果有悖常理的问题,提出一种基于证据信任因子的新的融合方法。根据证据的信任因子赋予其相应的权重,加权平均后得到期望证据,再进行合成。实验结果表明,该改进方法消除了高冲突度证据对合成结果的影响,具有更高的目标识别准确度。

When using classical D-S evidence theory,the basic probability assignment(BPA)of key parameters is often obtained by subjective experience,leading to the problem of low credibility of decision-making.A method to obtain the basic probability assignment by constructing BP neural network was proposed.The method utilized the powerful self-learning and non-linear mapping ability of BP neural network to normalize the output value to get the basic probability assignment.At the same time,in order to solve the paradoxical problem of synthetic evidence with high conflict degree,a new fusion method based on evidence trust factor was proposed.According to the trust factor of the evidence,giving the corresponding weight,the expected evidence was obtained after weighted average.Then fused expected evidence by D-S fusion formula.The experimental results showed that the proposed method could eliminate the influence of high conflict evidence on the synthesis results,and had a higher accuracy of target recognition.