【作者】 何峻；

【导师】 郭桂蓉； 付强；

【作者基本信息】 国防科学技术大学 ， 信息与通信工程， 2009， 博士

【摘要】 自动目标识别(Automatic Target Recognition,ATR)作为武器装备智能化的核心技术之一,为信息化战争中的目标探测、侦察监视和精确制导提供了有力支持,因而具有广泛的军事应用前景。论文将ATR评估(evaluation)定义为以ATR作为评价对象的行为活动。ATR评估能够为ATR技术的改进提供决策依据,并贯穿其整个研制过程,对促进ATR技术的快速发展具有重要意义。论文围绕ATR技术发展中的现实问题,在概率型指标估计与比较、多指标综合评价和扩展工作条件下技术效率与因素作用测算这三个方面深入开展了评估方法的研究,取得了一定成果。论文的主要内容如下:第二章研究ATR评估中最为基础的概率型单指标评估问题,按评估目的分为估计和比较两种情况来讨论。现有估计方法中普遍缺乏测试样本容量的预先分析,针对此情况,基于贝叶斯理论研究了概率型指标区间估计的要素,确定了区间估计方法和样本容量计算准则,定量分析了估计精度与样本容量的关系,以图表形式给出了一些典型估计精度要求下的最小样本容量,并且讨论了实际情况中的样本需求递减效应。针对现有比较方法中缺乏结论可信程度分析的现状,基于不确定推理提出了一种新的概率型指标比较方法,定量分析了典型情况下比较结果置信度与测试样本容量的关系,并且运用该方法揭示了经验做法中隐含的最大似然原理。第三章研究更具一般性的多指标综合评估问题,从决策分析角度提出了几种新的评估方法。针对ATR评估中的区间数多属性决策问题,基于分值模型提出了区间加权法和区间TOPSIS法,得到区间数形式的综合评分值,这有助于ATR评估中的柔性决策。针对ATR评估中的混合型多属性决策问题,基于关系模型提出了偏好矩阵法和次序关系法,实现了同时包含实数型、区间型和风险型三类指标的综合比较与排序。对于所提出的这几种多指标评估方法,还分别进行了实例分析。第四章研究扩展工作条件下的多指标评估问题,从效率测算角度重新考察了工作条件可变的ATR评估问题。首先针对扩展性及量测性评估方法在实际应用中遇到的困难,基于DEA理论提出了一种ATR技术效率分析方法,研究了求解过程中的技术细节,并结合实例来说明如何分析评估对象的ATR技术效率。然后根据效率分析的思想,针对以往采用性能建模方式的局限,基于Malmquist指数提出了一种非参数的因素作用测算方法,结合课题背景研究了评价指数的计算、分解等问题,并通过评估实例阐明如何度量扩展工作条件下因素变化的影响。论文研究虽然是结合了以雷达ATR为技术背景的多个重点科研项目,但是上述评估方法也可以推广到红外、激光和多传感器等ATR技术背景中。更多还原

【Abstract】 Automatic target recognition (ATR) is one of crucial technologies of intelligence weapon system. ATR gives the opportunity for target detection, surveillance, reconnaissance and precision attacking in the information warfare, so ATR has a broad military application. This dissertation defines ATR evaluation as the behaviors of assessing an ATR. As an important step in ATR development, ATR evaluation provides the decision foundation for its perfection, which is significantly valuable to accelerate the technology development. The dissertation focuses on some practical problems of ATR development, including the probability measure estimation and comparison, the multiple measures comprehensive evaluation, and the technical efficiency and factor influence calculation in extended operation condition (EOC). The main contributions of the dissertation are demonstrated as follows:In Chapter 2, two fundamental problems of ATR evaluation which based on a single probability measure are investigated. The estimation method and comparison method are discussed separately according to different evaluation purposes. To the question of lacking prior sample size analysis in the existing estimation methods, the elements of probability measure interval estimation based on Bayesian approach are analyzed. The interval estimation method and the corresponding criteria for sample size calculation are established and then be used to draw the relationship between the estimation precision and the sample size. Using the interval estimation method, the requirements of the minimum sample size for some typical estimation precision are given out with figure or table form, and as its application, the phenomenon of sample size descending in the real testing are also discussed. To the question of lacking confidence analysis in the existing comparison method, a new probability measure comparison method based on uncertainty inference is proposed. Using this new comparison method, the relationship between the confidence of comparison result and the requirement of test sample size is analyzed quantitatively, and as its application, the maximum likelihood principle within the experiential approach is revealed.In Chapter 3, the more general multiple measures comprehensive evaluation problems are investigated, and some new evaluation method are proposed in the perspective of decision making. To solve the interval multiple attribute decision making (MADM) problem in ATR evaluation, the interval weighted summation method and the interval TOPSIS method are proposed based on score model. The final interval comprehensive score conduces to a flexible decision for ATR evaluation. To solve the hybrid MADM problem in ATR evaluation, the preference matrix method and the order relation method are proposed based on relational model, which can rank and assess the evaluation objects with real, interval and random measures at the same time. As the illustration of these evaluation methods above, some application examples are also given.In Chapter 4, the multiple measures evaluation problems in EOC are investigated. The ATR evaluation concepts in variable operation conditions are surveyed in the perspective of efficiency measurement. Considering the practical difficulties of the extensibility and the scalability evaluation methods, a technical efficiency analytical method based on the data employment analysis (DEA) is proposed firstly. The details of its evaluation procedure are particularly discussed, and an application example is also given to illustrate how to calculate the technical efficiency of an ATR. Then for the sake of overcoming the shortages of performance modeling pattern in factor influence analyzing, a non parametric factor influence measurement method based on Malmquist index is proposed. The calculation and decomposition details of this evaluation index are discussed according to ATR evaluation background, and the application example is also given to demonstrate how to calculate the influence of a factor in EOC. Although the work of the dissertation is associated with some advanced scientific research programs which based on radar ATR technology, these evaluation methods can also be applied to other technical backgrounds, such as the infrared ATR, laser ATR, multi-sensor ATR and so on.更多还原