【作者】 邓翔宇；

【导师】 马义德；

【作者基本信息】 兰州大学 ， 无线电物理， 2013， 博士

【摘要】 PCNN(脉冲耦合神经网络)模型由于其神经元的同步脉冲发放特性和捕获特性,被广泛应用于数字图像分割、边缘检测、噪声滤除、图像增强、检索等处理,成为生物医学影像分析、遥感遥测图像处理、军事目标检测、交通图像处理等领域研究热点。在不断改善PCNN图像处理特性的同时,也存在着一些关键性难点成为国内外研究者共同关注的问题,其中PCNN模型自身数学特性的分析、模型参数的自适应设置对神经元特性的影响等是目前研究的难点和热点。本文从PCNN模型的系统方程出发,利用数学迭代法和离散系统分析法,对无耦合连接和耦合连接两种状态下的PCNN模型进行神经元点火机理分析,给出神经元点火时刻的数学表达式及其修正公式,同时揭示了PCNN模型本身的数学耦合特性——点火阶梯特性,以及其对网络生物学特性造成的覆盖现象,并对这种现象产生的机理和消除方法进行了分析,在此基础上分析了神经元点火特性受邻域神经元点火状态影响的情况,给出神经元点火时刻提前量数学表达式,进一步提出基于消除PCNN模型数学耦合特性和最大灰度值最小时刻点火两个约束条件的PCNN参数自适应设置方法。在完成神经元点火时刻受网络参数以及邻域神经元点火状态影响规律研究的基础上,本文提出一种PCNN的改进模型,并将其分别用于图像分割和噪声滤波中,同时也提出了一种基于PCNN的图像边缘检测新模型,并针对这三种图像处理,分别进行了参数自适应设置的分析和讨论。论文的主要工作如下：1.通过对PCNN模型离散方程的数学分析,给出了无耦合连接和耦合连接时神经元点火时刻的数学表达式及其修正公式,揭示了神经元理论点火时刻与实际点火时刻不一致的现象,提出了“数学耦合点火”和“点火阶梯图”的概念,并对神经元点火阶梯特性与网络参数之间的关系进行了详细分析；2.分别从本神经元点火时刻受点火阶梯的影响以及邻域神经元对本神经元的耦合作用受点火阶梯的影响两个角度,分析了PCNN模型数学耦合特性对网络生物学特性的影响情况,详细讨论了这种干扰和影响产生的机理,提出了消除模型数学耦合特性的参数自适应设置方法,并对算法性能进行了分析；3.对PCNN模型进行参数的规整,进一步对模型中的调制子系统进行调幅过程的模拟,分析了调制子系统参数设置对网络脉冲发放特性的影响情况,提出了一种具有“精细”脉冲发放特性的改进PCNN模型,并对改进模型的特性进行了分析,提出了参数自适应设置的方法；4.将上面提出的具有“精细”脉冲发放特性的改进PCNN模型应用于图像分割和噪声滤波中,针对于不同的应用完成了参数自适应设置的讨论,同时也提出了一种基于PCNN的图像边缘检测新模型,并给出了参数自适应设置的方法。更多还原

【Abstract】 With features of neuron synchronous pulse burst and capture, the Pulse Coupled Neural Network (PCNN) model has been widely applied in processing areas like digital image segmentation, edge detection, noise filtering, image enhancement, retrieval etc. and it has been the hot research area in the field of biomedical image analysis, remote sensing image processing, military target detection and traffic image processing. Yet features of PCNN image processing have been improved continually, there are still some key technical difficulties that has roused attentions from researchers home and abroad. Difficulties like mathematical properties of PCNN model itself and effects of parameters adaptive setting of the model on neuron have become the difficult and hot research subject.Guided by the system equation of PCNN model and by using mathematical iteration method and the discrete system analysis method neuron firing mechanism of PCNN model was analyzed under two kinds of state, coupling and without coupling. This paper presented the mathematical expression and modifier formula of neuron firing time, revealed the mathematical coupling feature of PCNN model itself, the step firing and the coverage influence it had on the biological characteristics of PCNN. The mechanism caused by that phenomenon as well as ways to eliminate have been analyzed. And based on what have been researched, this paper analyzed the situation that the neuron firing was influenced by the status of neighborhood neuron firing, brought out the mathematical expression of the lead of neuron firing timing and further put out the parameter adaptive setting method of PCNN based on the idea of eliminating the mathematical coupling feature of PCNN model and with two restrictions:the max gray and the minimum time firing. Based on the study of the law that neuron firing time is affected by network parameters and neighborhood neuron firing status, the writer presented an improved PCNN model and applied it into areas like image segmentation and noise filtering and a new model based on image edge detection of PCNN was also presented. In terms of those three ways of image processing, parameters adaptive setting of them have been analyzed and discussed separately. The main work as follows:1. By mathematical analysis of the discrete equation of PCNN model, the mathematical expression and modifier formula of neuron firing time under two kinds of state, coupling and without coupling were given. The research revealed the phenomenon that the theoretic neuron firing time disaccord with the actual neuron firing time, put out concepts of "mathematics firing" and "diagram of step firing". The relationship between step features of neuron firing and network parameter was analyzed in detail.2. From perspectives that neuron firing time itself as well as effects of neighborhood neuron on coupling effect of neuron themselves are influenced by firing step, the writer analyzed effects of the mathematical coupling feature of the PCNN model on the feature of network biology, discussed the mechanism caused by interference and influence in detail, put out the parameter adaptive setting method based on the idea of eliminating the mathematical coupling feature of PCNN model and the algorithm performance was analyzed too.3. Parameters of PCNN model was put in order and simulation for the process of modulation of subsystem of the model was analyzed and from which the information that how did parameters determination of the subsystem influence network pulse burst feature was given. Then an improved PCNN model with a "fine" pulse burst feature was put forward. Features of the improved model were analyzed and parameter adaptive setting method was provided.4. By applying the improved PCNN model with a "fine" pulse burst feature, mentioned above, into areas of image segmentation and noise filtering, parameters adaptive setting in different applications were discussed and a new image edge detection model based on PCNN as well as parameters adaptive setting methods were put out.更多还原