【作者】 刘健；

【导师】 张振川；

【作者基本信息】 东北大学 ， 计算机技术， 2008， 硕士

【摘要】 高技术条件下局部战争,是在化学威胁条件下进行的,作战进程空前加快,情况变化急剧,作战信息量骤然增加,要求作为作战保障兵种的防化兵作战指挥更迅速、更及时、更科学。神经网络识毒根据系统的毒剂征候,利用神经网络的非线性映射能力或模式识别功能,识别出系统的毒剂。从而进一步判定染毒浓度和范围,辅助指挥员优化作战行动方案,科学、快速地了解目标地区受化学攻击的状况,指导部队及时采取防护行动,最大限度降低毒剂的侵害。本论文围绕神经网络识毒建模及其在防化指挥自动化系统中的应用,着重探讨和研究了以下内容：阐述了神经网络识毒的概念与意义。从国内外研究神经网络识毒动态和我军防化指挥自动化系统建设现状入手,从必要性和可行性分析、总体需求分析,战术技术指标要求、功能要求几个方面论述了防化指挥自动化系统对神经网络识毒模型的需求。建立了神经网络识毒模型。包括神经网络与毒剂识别相关知识的介绍、神经网络识毒的基本方法、建立模型的数学依据、算法和数据库设计,为神经网络识毒模型在防化指挥自动化系统中的应用做好了前期准备工作。论述了特征提取与选择的概念和意义。提出了基于样本综合离散度和参数识别置信度两种识别参数选择的方法,为神经网络识毒打下了基础。从基于性能分析和基于小波分析进行了毒剂特征提取与选择。剖析了神经网络识毒的实现。进行了单个神经网络毒剂识别方法分析,提出了集成神经网络毒剂识别策略和神经网络识毒的软件构成,进行了神经网络识毒的仿真。测试结果表明：神经网络对毒剂有较好的识别能力。提出了神经网络识毒模型在防化指挥自动化系统中的应用构想。包括神经网络识毒模型在防化指挥自动化系统中的应用思路与要求、结构设计与应用方法。探讨了神经网络识毒信息在防化指挥自动化系统中的融合。进行了神经网络识毒信息流程分析和给出了查明毒剂种类的综合决策程序与方法,列举了神经网络识毒模型在防化指挥自动化系统中的应用示例。本论文构筑于Windows XP平台,采用MATLAB6.5、Delphi7.0和SQL Server2000等软件开发了神经网络识毒系统,通过检验正确识别率达96%以上,基本符合防化兵侦毒作战要求。论文指出了基于毒剂性能分析的特征提取与选择神经网络识别毒剂,适用于毒剂识别辅助决策,支持防化指挥自动化系统的指挥决策分系统；防化兵使用与化学传感器相联结的神经网络作战,是实现我军化学侦察自动化、智能化的一种有效方法。更多还原

【Abstract】 This essay surrounds the building model as well as application of the Neural Networks Distinguishing Chemical Agents(NNDCA) model in the Automation Command System of Chemical Defense(ACSCD), the main contents are listed as follows.-It setts forth the concept and significance of NNDCA. It starts with the domestic and overseas current researching situation of NNDCA and the condition of our corps construction of ACSCD, and discusses the need of ACSCD for NNDCA model from the necessity and viability assessment, the total need analysis, the military tactics and technique index sign requests, the function request, etc.This part introduces relating knowledge of the neural networks and distinguishing chemical agents, the basic method of NNDCA, the mathematics basis of building NNDCA model, the algorithm and database design. It prepares for the application of NNDCA model in ACSCD.The concept and meanings of feature extraction and selection are discussed. It believes parameters for distinguishing are also selected based on two different criteria:the cluster divergence of sample datas and the distinguishing reliability of parameter candidates, which be the foundation for NNDCA. It does chemical agent feature extraction and selectio from analysis base on the chemical agents characteristic and wavelet.This essay analyses the realization of NNDCA including single neural network distinguishing chemical agents. It puts forward to the strategy of the integrated neural networks distinguishing chemical agents and the software composing of NNDCA, carries on simulation of NNDCA.It puts forward the applied speculation of NNDCA model in ACSCD including the applied way of thinking and requesting, the structure designs and applied methods.It dissusses the fusion of the information of NNDCA in ACSCD. It analyses the information process of NNDCA, gives the comprehensive decision procedure and methods to distinguish the category of the chemical agents, and lists the applied example of NNDCA model in ACSCD.This essay points out that NNDCA through the feature extraction and selection base on chemical agents characteristic is fit for aid decision, this method supports the command decision-making subsystem. Our corps of chemical defense do battle with the neural networks combined with chemical sensor, which is a kind of valid method to realize the automation and intelligence in chemical reconnaissance.更多还原