海战场多传感器目标识别中神经网络的应用研究  

Research of Sea Battlefield Multi-sensor Target Recognition Based on LMBP Neutral Network

作　　者：刘楠楠[1] 张永利[1] 宋鹏汉 LIU Nan-nan;ZHANG Yong-li;SONG Peng-han(China Academy of Electronics and Information Technology,Beijing 100041,China)

机构地区：[1]中国电子科学研究院,北京100041

出　　处：《中国电子科学研究院学报》2020年第5期427-434,共8页Journal of China Academy of Electronics and Information Technology

摘　　要：由于海面目标所处的海战场环境的复杂性,以及海面目标探测获取的状态数据的不确定、缺失、模糊以及动态变化等,使得海面目标的综合识别非常困难。为了解决这些问题,需要对海战场多传感器目标综合识别中采用智能的神经网络进行应用研究。文中提出一种特征级采用LMBP神经网络算法,从多维度学习训练多传感器获取的数据,提取更多相关细节的目标特征属性信息,决策级采用证据理论的智能算法模型,即列文伯格误差反向传播结合证据理论算法的目标综合识别算法(Levenberg Marquardt Back Propagation-Dempster Shafer,简称LMBP-DS算法);然后通过Matlab仿真实验,比较不同的隐藏节点数对识别率的影响,找到LMBP-DS算法最佳的神经网络结构;通过对比实验得出:LMBP-DS算法比动量自适应学习BP-DS算法具有更快的收敛速度,同时具有更稳健高效的识别正确率,从而更适用于海战场多传感器目标综合识别。Because of the complexity of the sea battlefield environment where the sea targets are located,as well as the uncertainty,missing,fuzzy and dynamic changes of the state data obtained from the sea target detection,the comprehensive recognition of the sea targets is very difficult.In order to solve these problems,it is necessary to study the application of intelligent neural network in the sea battlefield multisensor target recognition.In this paper,a feature level algorithm based on LMBP neural network is proposed to learn and train the data obtained by multi-dimensional multi-sensor,and extract more detailed target feature attribute information.In the decision-making level,the intelligent algorithm model of evidence theory is used,that is,the Levenberg Marquardt Back Propagation Dempster algorithm Shafer,which is called LMBP-DS algorithm for short.Then through the MATLAB simulation experiment,the influence of different hidden nodes on the recognition rate is compared,and the optimal neural network structure of LMBP-DS algorithm is found.Finally,through the comparative experiment,it is concluded that LMBP-DS algorithm has faster convergence speed than momentum adaptive learning BP-DS algorithm,and has more robust and efficient recognition accuracy,so it is more applicable Comprehensive recognition of multi-sensor targets in sea battlefield.

关 键 词：海战场 目标综合识别 神经网络 动量-自适应学习速率 LMBP-DS 

分 类 号：TN97[电子电信—信号与信息处理] U666.7[电子电信—信息与通信工程]