基于聚类算法的红细胞形态分析与研究

发布时间：2018-05-22 14:31

本文选题：红细胞分类 + 数学建模　；参考：《湘潭大学》2017年硕士论文

【摘要】：随着经济的发展,人们的生活质量和文化水平不断提升,大家越来越重视自身及家人的健康问题,人体健康情况与红细胞的形态和数量息息相关,许多疾病的诊断必须依靠红细胞形态与数量作为依据。本文根据红细胞自动识别系统的作业要求,运用图像处理及模式识别技术,对其中的关键技术进行研究,用实验进行论证,较好地解决了红细胞分类识别的问题。本文针对红细胞显微图像自动识别系统的研究课题,在前人工作的基础上,对红细胞语义模型的建立、红细胞图像分割算法、红细胞特征提取与选择、模糊聚类分类识别等方面进行研究和实验,主要完成以下工作:1、在图像分割研究方面,在分析红细胞图像特点及部分分割算法后,提出了一种基于图割的红细胞分割算法,此算法是通过迭代的方式将目标从一个复杂的背景中提取出来,解决了无关区域和内部空洞对分割的影响和多粘连细胞的分割效果差的问题,减少了算法的复杂性,提高了分割速度和分割准确率。2、在形状特征提取中,研究了各类异形红细胞的基本形态,然后逐类建立语义模型,根据语义模型确定模型特征的数学表述方法,提出了边缘突起点和凹点两个新的形状特征,然后对其他形状特征进行改善和选择,为进一步识别各种异形红细胞提供了依据,并且提高了运行效率和分类准确率。3、在纹理特征提取中,采用共生矩阵分析法描述红细胞的纹理特征,经过实验数据的对比,选取了基于共生矩阵的二阶矩、对比度、相关性、熵、逆差矩、和熵、均值差、差熵、方差和9个纹理特征。4、学习了模糊聚类的基本原理,研究了FCM聚类算法,针对其算法在模糊距离计算方面的不足进行了改进,提高了算法的速度和准确度。采用改进的FCM算法对11类红细胞进行聚类识别,对聚类结果进行了分析。实验证明采用改进的聚类算法实现了代码的高效运行,提高了图像数据处理的效率,满足图像实时处理的工程需求,完成了对红细胞的分类,得到较好的分类效果。
[Abstract]:With the development of economy, people's quality of life and educational level are improving constantly. People pay more and more attention to the health problems of themselves and their families. Human health is closely related to the shape and quantity of red blood cells. The diagnosis of many diseases must depend on the morphology and quantity of red blood cells. According to the operational requirements of the red blood cell automatic recognition system, this paper studies the key technology by using image processing and pattern recognition technology, and proves it by experiment, which solves the problem of red blood cell classification and recognition. In this paper, aiming at the research topic of erythrocyte microscopic image automatic recognition system, on the basis of previous work, the establishment of erythrocyte semantic model, red cell image segmentation algorithm, red blood cell feature extraction and selection, In the aspect of image segmentation, after analyzing the characteristics of red blood cell image and partial segmentation algorithm, a red blood cell segmentation algorithm based on graph cutting is proposed. The algorithm extracts the target from a complex background by iterative method, solves the problem of the influence of irrelevant region and internal cavity on segmentation and the poor segmentation effect of multi-adhesion cells, and reduces the complexity of the algorithm. The segmentation speed and segmentation accuracy are improved. In shape feature extraction, the basic morphology of all kinds of special-shaped red blood cells is studied, and then the semantic model is established one by one, and the mathematical expression method of the model feature is determined according to the semantic model. Two new shape features, starting point of edge process and concave point, are proposed, and other shape features are improved and selected, which provides a basis for further recognition of various special-shaped red blood cells. In texture feature extraction, the co-occurrence matrix analysis method is used to describe the texture features of red blood cells. Through the comparison of experimental data, the second moment, contrast and correlation based on co-occurrence matrix are selected. Entropy, deficit moment, and entropy, mean difference, difference entropy, variance and nine texture features. 4. The basic principle of fuzzy clustering is studied, and the FCM clustering algorithm is studied. The speed and accuracy of the algorithm are improved. The improved FCM algorithm is used to identify 11 red blood cells and the clustering results are analyzed. The experimental results show that the improved clustering algorithm can efficiently run the code, improve the efficiency of image data processing, meet the engineering requirements of real-time image processing, and complete the classification of red blood cells.
【学位授予单位】：湘潭大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R446.1;TP391.41

【参考文献】