【作者】 谭枫；

【导师】 杨莘元；

【作者基本信息】 哈尔滨工程大学 ， 通信与信息系统， 2010， 博士

【摘要】 文化算法是一种起源于文化进化过程的进化计算模型,它的主要思想是从进化种群中获取求解问题的知识并用于指导搜索过程。与其它传统的进化算法不同,文化算法是一种具有种群空间和信念空间的双层遗传系统,能够以显性的方式获取、存储和整合问题求解过程中的经验知识。因此,它的进化速度超越了单纯依靠生物基因遗传的生物进化速度。这一特点引起了研究者们的广泛关注,目前文化算法的理论研究工作尚处于起步阶段。本文在讨论文化算法优化理论的基础上,对文化算法的基本框架和算法进行了改进,并将其应用于图像阈值分割中。主要研究内容及创新点如下：首先,针对无约束优化中易陷入局部最优的问题,引入了介入式文化算法框架,并将该框架应用于无约束优化问题,提出了两种介入式文化算法,即具有双通信协议的介入式文化算法和介入式文化粒子群算法。具有双通信协议的介入式文化算法以基本文化算法为基础,以进化规划为种群模型。利用信念空间中的形势知识判断算法是否陷入局部最优,通过介入操作来保持种群空间的多样性。介入式文化粒子群算法以粒子群优化算法为种群模型,利用全局最优解判断算法是否陷入局部最优,通过r/K选择策略来保持种群的多样性。与基本的文化算法相比,所提出的具有双通信协议的介入式文化算法能够适应更多的无约束优化函数,而介入式文化粒子群算法更适合高维复杂函数的优化。其次,针对约束优化问题,对介入式文化算法框架进行了调整,提出了两种介入式文化算法,即基于分层结构模型的介入式文化算法和基于动态模型的介入式文化算法。基于分层结构模型的介入式文化算法以进化规划为种群模型,以树状区域图示为信念空间模型。利用信念元的基本属性判断算法是否陷入局部最优,通过介入机制直接操作信念空间,以达到跳出局部最优的目的。基于动态模型的介入式文化算法以进化规划为种群模型,以动态函数优化中文化算法的知识结构为信念空间模型。利用历史知识判断算法是否陷入局部最优,直接操作信念空间的相关知识协助算法跳出局部最优。与基于分层结构模型的文化算法相比,所提出的基于分层结构模型的介入式文化算法的优化性能更好。与改进的进化规划相比,所提出的基于动态模型的介入式文化算法的成功率更高、稳定性更好。与其它具有竞争力的约束优化算法相比,所提出的基于动态模型的介入式文化算法精度更高、收敛速度更快。最后,将介入式文化粒子群算法应用于图像阈值分割中,提出了基于介入式文化粒子群算法的多阈值图像分割法和基于介入式文化粒子群算法的二维图像分割法。在多阈值分割中,分别将介入式文化粒子群算法与最大类间差和最大熵法相结合,对图像进行双阈值和三阈值分割。在二维阈值分割中,将介入式文化粒子群算法与二维最大类间差相结合,对图像进行二维阈值分割。与同类型的图像阂值分割方法相比,本文提出的方法具有更高的分割精度、更快的分割速度。更多还原

【Abstract】 Cultural algorithm (CA) is an evolutionary computation model originated from culture evolutionary process. Its core is to explicitly acquire problem-solving knowledge from the evolving population and in return to apply the knowledge to guide the search. Different from traditional evolution algorithms, CA is a dual inheritance system that models two levels of evolution:the population space and the belief space, and CA can provide an explicit mechanism for acquisition, storage and integration of problem solving experience. As a result, this evolutionary speed surpasses the speed of the single biological genetic evolution. The scholars and the researchers have been paying more and more attention to this particular feature, although CA’s theoretical research is still in the initial stage.Based on the discussion of CA’s theory, this paper improves CA’s framework and algorithms, and applies the framework into the image segmentation. The main research work and the creativities in the dissertation are as follows.Firstly, as far as the question easily-trapped in local optimum in the unconstrained optimization is concerned, the intervention cultural algorithm framework is introduced, and furthermore, apply this framework to unconstrained optimization, in order to propose two types of intervention cultural algorithms-intervention cultural algorithm with bi-communication protocol (ICAEP-bcp) and the intervention cultural particle swarm optimization (ICA-PSO).ICAEP-bcp is based on the basic CA and the evolutionary programming is used as the population model. The research applies the situational knowledge in the belief space to judge whether the algorithm is trapped in local optimum, and uses the intervention operation to keep the variety in the population space. ICA-PSO takes the particle swarm optimization as the population model, using the global optimum to judge whether the algorithm is trapped in the local optimum, through r/K selection strategies to keep the variety of the population. Compared with the basic CA, here the proposed ICAEP-bcp can fit the unconstrained optimization functions more, while ICA-PSO can fit high-dimensional complicated function’s optimization more.Secondly, according to the questions of the constrained optimization, to modify the intervention CA’s framework, two intervention cultural algorithms are proposed. One is based on the hierarchical architecture model (IHAM),the other is based on the dynamic model (ICAEP-DM).IHAM takes evolutionary programming as the population model, and takes tree regional structure as the belief space model. The basic characteristics of the belief cell are used to judge whether the algorithm is trapped in local optimum, through the intervention mechanism to operate the belief cell, in order to jump out of the local optimum. ICAEP-DM takes the evolutionary programming as the population model, and takes the knowledge structure in dynamic model as the belief space model. Applying the historical knowledge to judge whether the algorithm is trapped in local optimum, the knowledge concerning the belief space is operated to help the algorithm to jump out of the local optimum. Compared with the CA based on hierarchical architecture model, IHAM owns priorities. Compared with the improved evolutionary programming, the proposed ICAEP-DM has high success ratio and better stability. Compared with other competitive constraint optimization algorithm, ICAEP-DM has higher accuracy and faster convergence.Finally, the idea of ICA-PSO is introduced into the image segmentation. Combining the Otsu’s method and maximum entropy method respectively, the multi-threshold image segmentation methods based on ICA-PSO are proposed for double thresholding and three thresholding. Combing the two-dimensional Otsu’s method, the two-dimensional segmentation method based on ICA-PSO is proposed. Compared with other methods, the proposed methods in this paper give better performance and provide higher convergence speed.更多还原