【作者】 骆祥峰；

【导师】 高隽；

【作者基本信息】 合肥工业大学 ， 计算机应用技术， 2003， 博士

【摘要】 认知图是计算智能研究的一个热点，它是一种定性推理技术，其提供了一个有效的软计算工具来支持基于先验知识的自适应行为，但是认知图缺少学习及自适应能力，是一个封闭的系统。图像分析与理解是目前国内外的一个热点与难点问题，对它们的研究方法较多，但很多方法较少同时考虑图像物体与3D场景中的先验知识及计算智能。本文针对认知图理论及图像分析与理解中存在的问题，着重对认知图的学习、知识的自动获取与模型的扩展这些理论问题进行研究，积极探索如何应用认知图来解决目前图像分析与理解方法中存在的某些缺陷。 对于认知图理论的研究：本文在对认知图的国内外研究现状作了深入分析的基础上，首先对认知图的稳定性进行尝试性的讨论。对于认知图的学习，提出不平衡度的概念来度量认知图与真实世界存在的差异。对于认知图知识的自动获取，提出概型自适应环境变化的原理。对于复杂环境下认知图的相互合作与协调的不一致性，提出依据欲望概念与威胁概念节点状态值来使其协调一致的思想。 本文把条件概率、不确定性理论及知识库引入认知图中，提出“概率模糊认知图”、“基于信任知识库的概率模糊认知图”及“扩展动态认知网络”来表示事物间因果关系测度的不确定性、因果联系的时空特性及专家对知识的不确定性，从而扩展了认知图模拟现实世界的能力。本文也对概念间具有因果时间延迟及非因果关系认知图的推理进行了初步的讨论。 对于认知图在图像分析及理解中的应用：本文在认知图理论研究的基础上，把认知图应用到图像分析与图像理解中。实现了一种基于模糊认知图的线特征检测与形成方法、基于模糊认知图的基本形状识别方法、基于概率模糊认知图的目标识别方法及基于认知图的图像理解方法，并将图像理解与认知图应用到机器人的高层规划系统中。这些方法在一定程度上克服其他方法存在的某些缺陷，对发展新型的图像分析与理解方法具有一定的启迪作用。更多还原

【Abstract】 Cognitive map is one of the hotspots in computing intelligence area. It is a qualitative inference technology, which provides a tool of soft computation to support self- adaptive behaviors that based on prior knowledge. However, cognitive map lacks of the capability of learning and self- adaptive. Image analysis and understanding are hotspots and difficult problems in image areas. Many methods have been proposed in research on image analysis and understanding, but few of them considered prior knowledge and computing intelligence simultaneously. Due to above problems, this dissertation discussed theory of cognitive map and its application in image analysis and understanding, whose mainly research is on learning of cognitive map, auto-knowledge acquiring of cognitive map and exploring cognitive map applications in image analysis and understanding, expecting to solve some drawbacks in the previous methods of image analysis and understanding.About the cognitive map theory: The stability of cognitive map is discussed firstly. For the study capability of cognitive map, objective datum is emphasized and the degree of unequilibrium is proposed to measure the difference between the cognitive map and the real world, so the expert knowledge and the objective datum can be unified in cognitive map.According to the values of the desire and theater concepts, a new idea is proposed which can acquire the knowledge of real world automatically.For the extend model of cognitive map, conditional probability, theory of uncertainty and knowledge database are introduced to cognitive map, and fuzzy cognitive map (FCM), probabilistic fuzzy cognitive map (PFCM), belief knowledge database based probabilistic fuzzy cognitive map (BKPFCM), "extended dynamic cognitive network" are presented. Therefore, those extended models can express the fuzzy and belief measure of uncertainty causal relationships and expert knowledge with uncertainty. Finally, the simulating capability of cognitive map is extended naturally.About the applications of cognitive map: Line feature detection and form based on fuzzy cognitive map, basic shapes recognition based on fuzzy cognitive map, object recognition based on probabilistic fuzzy cognitive map, a new method of image understanding based on cognitive map and robot high-level planning based on fuzzy cognitive map have been realized in this dissertation. Those methods can also go over some defects in previous methods of image analysis and understanding to a certain extent, which may be beneficial to developing some new methods in this field.更多还原