改进粒子群优化算法在眼底出血点分割中的研究

发布时间：2018-03-14 19:53

本文选题：粒子群优化算法　切入点：糖尿病视网膜疾病　出处：《哈尔滨工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：图像分割是特别重要的一种图像前处理手段,亦是机器视觉和人工智能领域的先决和根基工作之一。根据统计,由人类生活方式的影响,糖尿病发病的几率异常高,然则糖尿病后期引发的糖尿病视网膜疾病(Diabetic Retinopathy,DR)会引发眼底出现出血点。所以,经由图像分割技术检索眼底出血点可以高效的协助医生的诊疗,极大的程度上提升医生的工作效率。这些年来,有关图像分割的研究有很多,然而,有关图像分割的算法主要集中于图像全体区域分割和模板匹配分割。而眼底出血点特征非常弱、外形和规则不完全相同,因此,通过当前现有的图像分割算法不容易实现令医学应用领域满足可行的结果。而,当今有关眼底出血点分割的方法主要是依赖于典型机器学习和深度学习两个方面,然则典型机器学习和深度学习须要庞大的人工标识、特征抽取、调整参数,这样不单大量耗时并且成效比较局限。本文通过对DR和眼底出血点大量学习和研究,联合粒子群优化(Particle Swarm Optimization,PSO)算法展开研究,首先使用PSO算法寻找出血点的聚类中心和出血点所处的矩形位置,此后采用改进的活动轮廓模型(Active Contour Model,ACM)分割出眼底出血点的轮廓。本文做了三方向改进,第一:对于PSO算法容易陷入局部最优解的问题提出以Metropolis准则为基础的PSO算法;第二:关于眼底出血点确定方面,本文联合边缘、区域、外形特征构造一种特征相似系数并引入模糊聚类,构建一种新型PSO适应度函数,用于出血点聚类;第三:对于当今存在的ACM分割速度迟缓、对起始轮廓敏感的弊病提出以粒子群聚类为基础并加入聚类系数的局部ACM,让起始曲线更加贴近目标边界。通过Matlab仿真实验,本文改进的PSO和局部ACM在稳定性、速度、精度、抗击噪声能力方面获得了提升,此外,本文新提出的眼底出血点分割算法能够准确分割出出血点轮廓且在人工消耗、样本需求、分割精度等方面优于目前存在的基于机器学习眼底出血点检测算法。
[Abstract]:Image segmentation is a particularly important means of image preprocessing and one of the first and fundamental tasks in the field of machine vision and artificial intelligence. However, diabetic retinopathy (DRR) caused by late diabetic retinopathy can cause bleeding spots in the fundus. Therefore, retrieval of fundus bleeding points by image segmentation technology can effectively assist doctors in diagnosis and treatment. Greatly improve the efficiency of doctors. Over the years, there has been a lot of research on image segmentation, however, The algorithms of image segmentation mainly focus on the segmentation of all regions of the image and template matching segmentation. But the feature of eyeground bleeding point is very weak, and the shape and rules are not identical, so, The current image segmentation algorithms are not easy to achieve the feasible results in the field of medical application. However, the current segmentation methods of ocular fundus bleeding points mainly depend on two aspects: typical machine learning and deep learning. However, typical machine learning and deep learning require a large amount of manual identification, feature extraction, and adjustment of parameters, which are not only time-consuming but also limited in effectiveness. The Particle Swarm Optimization (PSO) algorithm is studied in this paper. Firstly, the PSO algorithm is used to find the cluster center of the bleeding point and the rectangular position of the bleeding point. After that, the active Contour Model (ACM) is used to segment the contour of the bleeding point in the eye fundus. In this paper, we improve the contour in three directions. Firstly, we propose a PSO algorithm based on the Metropolis criterion for the problem that the PSO algorithm is easy to fall into the local optimal solution. Second, regarding the determination of ocular fundus bleeding point, this paper combines the features of edge, region and shape to construct a characteristic similarity coefficient and introduce fuzzy clustering to construct a new PSO fitness function, which can be used for haemorrhage point clustering. Thirdly, for the slow speed of ACM segmentation, the defect of sensitivity to the initial contour is put forward, which is based on particle cluster and adding the clustering coefficient to make the starting curve closer to the target boundary. Through the Matlab simulation experiment, The improved PSO and local ACM have been improved in terms of stability, speed, accuracy and ability to resist noise. In addition, the new algorithm proposed in this paper can segment the contour of bleeding points accurately and consume them manually. The segmentation accuracy is superior to the existing machine-learning-based eyeground haemorrhage detection algorithm.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R587.2;R774.1;TP391.41

【相似文献】