混合优化算法及其在图像处理中的应用研究

发布时间：2018-07-20 11:43

【摘要】：复杂科学与工程中问题的计算量极大,常用的确定性最优化算法面对上述问题在有限的时间内经常会出现失效的情况。因此基于自然进化过程的进化计算得到了广泛的研究和关注,并在很多实际问题中得到了成功的应用。然而单一的进化算法或者群集智能算法或者本身都存在某些不足之处,需要进一步改善。多元混合算法是一类基于多种单一算法相互融合共同完成优化过程的算法,其优点包括平衡性好,组合灵活,鲁棒性强,适合复杂的优化问题。研究者们对许多混合算法进行了研究,取得了较好的效果。然而进化计算方法众多,相关理论和实践并未完善,值得进一步研究。本文根据常用进化算法和群集智能算法的特点,提出了一类新的算法混合模式,串行混合、并行混合和串并行混合,选取了6种常见的优化算法进行混合,并应用于图像处理优化求解问题中,主要工作如下:1.使用CEC2015中的15个测试函数对遗传算法、差分进化算法、粒子群算法、人工蜂群算法、杜鹃搜索算法、萤火虫算法这六个优化算法进行了测试,并结合算法的流程,归纳出算法的收敛性、搜索能力以及跳出局部最优等特性。2.提出了具体的混合策略:串行混合、并行混合和串并行混合。实现了六种串行混合算法和六种并行混合算法,并进行了仿真测试。结果表明混合算法的性能更加均衡,在优化复杂问题上有较好的效果。3.研究和实现了多元混合算法在图像分割,图像增强和图像匹配中的应用。选取了四个性能较好的混合算法以及对应的单一算法进行对比试验。选取的四个混合算法包括:串行粒子群杜鹃搜索算法、串行差分进化杜鹃搜索算法、并行差分进化遗传算法、并行粒子群差分进化算法;对应的单一算法包括:粒子群算法、杜鹃搜索算法、差分进化算法、遗传算法。试验结果表明混合优化算法在图像分割、图像增强以及图像匹配中能够快速得到较优的结果,且各算法在不同的图像上的处理效果相对稳定。总体来说,本文提出了一类最优化算法的混合模式,并根据该模式实现了数种混合算法且在图像分割、图像增强、图像匹配中进行了应用。实验结果表明使用该混合模式得到的混合算法继承了对应的单一算法的特性,对不同的优化问题有良好的优化性能和稳定性。
[Abstract]:The computational complexity of the problems in complex science and engineering is very large. The commonly used deterministic optimization algorithms often fail in the limited time in the face of these problems. Therefore, evolutionary computing based on natural evolutionary processes has been widely studied and paid attention to, and has been successfully applied in many practical problems. However, the single evolutionary algorithm or cluster intelligence algorithm or itself has some shortcomings, which need to be further improved. Multivariate hybrid algorithm is a kind of algorithm based on the fusion of a variety of single algorithms to complete the optimization process together. Its advantages include good balance, flexible combination, strong robustness, suitable for complex optimization problems. Researchers have studied many hybrid algorithms and achieved good results. However, there are many evolutionary computing methods, and the relevant theory and practice are not perfect, which is worthy of further study. According to the characteristics of common evolutionary algorithms and swarm intelligence algorithms, this paper presents a new class of hybrid algorithms, serial hybrid, parallel hybrid and series-parallel hybrid, and six common optimization algorithms are selected to mix. And applied to the image processing optimization problem, the main work is as follows: 1. This paper uses 15 test functions in CEC2015 to test the genetic algorithm, differential evolution algorithm, particle swarm optimization algorithm, artificial bee colony algorithm, rhododendron search algorithm and firefly algorithm. The convergence, searching ability and jumping out of local optimum of the algorithm are summarized. 2. 2. Specific mixing strategies are proposed: serial mixing, parallel mixing and series-parallel mixing. Six serial hybrid algorithms and six parallel hybrid algorithms are implemented and simulated. The results show that the performance of the hybrid algorithm is more balanced and has a better effect on the optimization of complex problems. The application of multivariate hybrid algorithm in image segmentation, image enhancement and image matching is studied and implemented. Four hybrid algorithms with good performance and the corresponding single algorithm are selected for comparative test. The four hybrid algorithms selected include: serial particle swarm cuckoo search algorithm, serial differential evolution rhododendron search algorithm, parallel differential evolution genetic algorithm, parallel particle swarm differential evolution algorithm; Rhododendron search algorithm, differential evolution algorithm, genetic algorithm. The experimental results show that the hybrid optimization algorithm can quickly obtain better results in image segmentation, image enhancement and image matching, and the processing effect of each algorithm on different images is relatively stable. In general, this paper proposes a kind of hybrid pattern of optimization algorithm, and implements several hybrid algorithms according to this pattern, and it is applied in image segmentation, image enhancement and image matching. The experimental results show that the hybrid algorithm inherits the characteristics of the corresponding single algorithm and has good optimization performance and stability for different optimization problems.
【学位授予单位】：湖北工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】