多源引导信息融合及其关键技术研究

发布时间：2018-01-17 22:20

本文关键词：多源引导信息融合及其关键技术研究　出处：《中国科学院研究生院(长春光学精密机械与物理研究所)》2016年博士论文　论文类型：学位论文

【摘要】：随着微电子技术、计算机技术、传感器技术的发展,多源信息融合技术已逐步发展为一个新的研究方向,并得到了广泛应用。本文针对多源引导信息融合的特点,在其模型框架中分析了多源引导信息融合的几个关键技术,重点研究了其中的目标识别和威胁估计两个方面,提出相应的方法,应用于某光电引导跟踪系统中。本文的主要工作和贡献如下:1.介绍了光电引导跟踪系统多源引导信息融合的基本理论和模型框架,对其中目标识别和威胁估计技术的研究情况进行了归纳总结;介绍了现有目标识别和威胁估计的相关算法,分析了各自的优缺点,并确定了本文所采用的主要技术手段。2.针对区间直觉模糊集和灰关联分析的特点,提出了一种基于区间直觉模糊灰关联分析的目标识别算法,应用于光电引导跟踪系统的目标识别过程中。该方法根据多源引导信息来获取目标特征参数,采用区间直觉模糊数来表征特征参数对目标类别的隶属度和非隶属度,形成目标识别矩阵。从中提取出正、负理想识别策略,采用灰色关联理论对识别矩阵进行分析,从而建立各目标类别的灰关联序。实验结果表明,基于区间直觉模糊灰关联的目标识别算法的识别误差小于采用单一方法的识别误差,并且能够避免组合爆炸问题。3.针对传统方法在比较直觉模糊值和区间直觉模糊值时的不足,提出了综合函数法,应用于对目标识别矩阵的分析过程中。该方法对得分函数和精确函数进行加权融合,并利用决策结果对加权因子进行自适应调节。综合函数法可以有效改善传统方法中由于净利润最大化而带来的缺陷,并对两模糊值之间的差值提供一个估计。4.针对粒子群优化算法(PSO)和BP神经网络的特点,对PSO算法进行了改进,提出了基于改进PSO算法优化BP神经网络(MPSO-BP)的目标威胁估计算法,应用于光电引导跟踪系统的目标威胁估计过程中。该方法采用PSO算法优化BP神经网络的初始权值和阈值,在PSO优化过程中引入变异过程,并对加速因子进行自适应处理,以避免粒子种群的快速趋同效应,使优化后的BP神经网络能够更好地预测输出。实验结果表明,MPSO-BP网络的预测误差小于BP网络和PSO-BP网络,并且能够较好地解决对于小样本训练问题,避免神经网络陷入局部极值。5.针对萤火虫优化算法(GSO)、磷虾群算法(KH)和支持向量机(SVM)的特点,建立了基于GSO算法优化SVM(MGSO-SVM)、以及KH算法优化SVM(KH-SVM)的目标威胁估计模型,并提出了基于各自模型的算法,应用于光电引导跟踪系统的目标威胁估计过程中,使优化后的SVM网络能够更好地完成目标威胁估计。实验结果表明,GSO-SVM和KH-SVM的预测误差明显小于SVM。
[Abstract]:With the development of microelectronic technology, computer technology and sensor technology, multi-source information fusion technology has gradually developed into a new research direction. And has been widely used. According to the characteristics of multi-source guided information fusion, this paper analyzes several key technologies of multi-source guided information fusion in the framework of its model. Two aspects of target identification and threat estimation are studied, and the corresponding methods are proposed. The main work and contributions of this paper are as follows: 1. The basic theory and model framework of multi-source guided information fusion are introduced. The research of target identification and threat estimation is summarized. This paper introduces the existing algorithms of target recognition and threat estimation, analyzes their advantages and disadvantages, and determines the main technical means used in this paper. 2. Aiming at the characteristics of interval intuitionistic fuzzy sets and grey association analysis. A target recognition algorithm based on interval intuitionistic fuzzy grey correlation analysis is proposed, which is applied to the target recognition process of photoelectric guided tracking system. The interval intuitionistic fuzzy number is used to represent the membership and non-membership degree of the feature parameters to the target category, and the target recognition matrix is formed, from which the positive and negative ideal recognition strategies are extracted. The grey correlation theory is used to analyze the recognition matrix, and the grey correlation order of each target category is established. The experimental results show that. The recognition error of the target recognition algorithm based on interval intuitionistic fuzzy grey correlation is smaller than that of the single method. And can avoid the combinatorial explosion problem. 3. Aiming at the shortcomings of traditional methods in comparing intuitionistic fuzzy values and interval intuitionistic fuzzy values, a comprehensive function method is proposed. The method is applied to the analysis of target recognition matrix, which combines the score function and the accurate function. Using the decision result to adjust the weighting factor adaptively, the comprehensive function method can effectively improve the defects caused by the net profit maximization in the traditional method. The difference between the two fuzzy values is estimated. 4. According to the characteristics of particle swarm optimization (PSO) and BP neural network, the PSO algorithm is improved. A target threat estimation algorithm based on improved PSO algorithm is proposed to optimize BP neural network. PSO algorithm is used to optimize the initial weight and threshold of BP neural network, and the mutation process is introduced in the process of PSO optimization. In order to avoid the fast convergence effect of particle population, the optimized BP neural network can predict the output better. The prediction error of MPSO-BP network is smaller than that of BP network and PSO-BP network, and it can solve the problem of small sample training. To avoid the neural network falling into local extremum. 5. Aiming at the characteristics of the firefly optimization algorithm (GSOO), krill swarm algorithm (KH) and support vector machine (SVM). The target threat estimation model based on GSO algorithm optimization and KH algorithm optimization is established, and the algorithm based on each model is proposed. In the process of target threat estimation in photoelectric guided tracking system, the optimized SVM network can accomplish the target threat estimation better. The prediction error of GSO-SVM and KH-SVM is obviously smaller than that of SVM.
【学位授予单位】：中国科学院研究生院(长春光学精密机械与物理研究所)
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP391.41

【相似文献】