遗传改进粒子群优化特征选择的研究与应用

发布时间：2018-05-25 02:16

本文选题：粒子群优化 + 遗传算法　；参考：《云南大学》2015年硕士论文

【摘要】：本文将遗传改进的离散的粒子群优化引入到特征选择中,并由此构建甲状腺结节恶性风险评估诊疗系统。本文主要包括以下内容：首先,对优化问题、粒子群优化、遗传算法和特征选择等概念作了概述。深入分析了粒子群优化,回顾了几种主要的粒子群优化改进算法。最后,介绍了特征选择的相关概念。然后,利用离散的粒子群优化具有天然编码的这一特性,将遗传算法的基本操作施用于离散的粒子群优化中,实现了基于遗传改进的离散的粒子群优化的核心算法,并对其算法性能进行了标准函数测试。根据课题目标,将该算法应用于特征选择问题中,并与其他几种主要的特征选择方法进行了对照试验。试验证明,本文提出的遗传改进的粒子群优化能明显地提升特征选择的寻优能力。根据上述工作,构建甲状腺结节恶性风险评估诊疗系统。在构建分类器之前,运用数字图像处理的技术对甲状腺结节的超声影像进行了特征提取,共提取出79个结节图像的形态和纹理特征。随后,将遗传改进的粒子群优化用于上述特征的选择。将得到的最优特征子集作为特征向量,用支持向量机对甲状腺结节的良恶性进行分类识别,训练得到的分类器精度可达到88.20%,性能超过了常规特征选择方法得到的同类分类器。根据研究结果可以看出,结节的紧致度、平滑度等在对甲状腺结节进行良恶性无创预判中起到关键作用。最后,在研究粒子群优化、遗传算法等进化计算算法的基础上,扩展到随机过程等相关的领域,并做了简要的叙述,以此描绘了未来工作的主要方向和领域。
[Abstract]:In this paper, the discrete particle swarm optimization based on genetic improvement is introduced to feature selection, and a diagnosis and treatment system for malignant risk assessment of thyroid nodules is constructed. This paper mainly includes the following contents: Firstly, the concepts of optimization problem, particle swarm optimization, genetic algorithm and feature selection are summarized. In this paper, particle swarm optimization (PSO) is deeply analyzed, and several improved PSO algorithms are reviewed. Finally, the concept of feature selection is introduced. Then, taking advantage of the natural coding property of discrete particle swarm optimization, the basic operation of genetic algorithm is applied to discrete particle swarm optimization, and the core algorithm of discrete particle swarm optimization based on genetic improvement is realized. The performance of the algorithm is tested by standard function. According to the objective of this paper, the algorithm is applied to the feature selection problem and compared with other major feature selection methods. The experiments show that the improved particle swarm optimization proposed in this paper can obviously improve the ability of feature selection. Based on the above work, the diagnosis and treatment system of thyroid nodule malignant risk assessment was constructed. Before the classifier was constructed, digital image processing technique was used to extract the features of thyroid nodule images, and 79 nodule images were extracted from the features of morphology and texture. Then, the genetic improved particle swarm optimization is applied to the selection of the above characteristics. Using the optimal feature subset as the feature vector, the support vector machine is used to classify the benign and malignant thyroid nodules. The trained classifier has a precision of 88.20, and its performance is better than the similar classifier obtained by the conventional feature selection method. According to the results, the tightness and smoothness of the nodules play a key role in the prediction of benign and malignant thyroid nodules. Finally, based on the research of particle swarm optimization (PSO) and genetic algorithm (GA), this paper extends to the related fields such as stochastic processes, and gives a brief description, which describes the main directions and fields of future work.
【学位授予单位】：云南大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：R581;TP18

【参考文献】