一种模糊聚类的遥感影像分析方法研究

The study of remote sensing image analysis method based on fuzzy ISODATA clustering

针对传统模糊聚类的遥感影像分析方法的不足,重点研究基于模糊ISODATA聚类的遥感影像分析。通过Matlab软件编程实现基于迭代自组织数据分析技术、模糊C均值聚类、模糊ISODATA算法对合成图像、纹理图像及真实遥感影像的分类,并对其分类结果进行讨论。通过实验数据对比,评价FISODATA算法的优越性。实验结果表明:ISODATA算法及FISODATA算法都能够实现变类,而FCM算法只能在固定聚类数下进行分类,但是,ISODATA算法分类机制不稳定,不能每次都确定正确聚类数。在迭代过程中,将FISODATA算法引入模糊集理论,便能够快速准确的实现聚类数的确定。

The traditional remote sensing image analysis method of fuzzy clustering has many disad- vantages. This paper focuses on the research of remote sensing image based on fuzzy ISODATA clustering analysis. Based on the Iterative Self-Organizing Data Analysis Techniques Algorithm（ISODATA）. Fuzzy C-means（FCM）. Fuzzy ISODATA（FISODATA）, which can achieve composite images＇, texture ima- ges＇ and real remote sensing images＇ classification. It discusses the result of the classification and evaluates the advantages of the algorithm FISODATA. Experimental results show that the ISODATA algorithm and the FISODATA algorithm can realize the image classification, and the FCM algorithm can classfy objects under the fixed clustering number. However, the mechanism of ISODATA algorithm is not stable and can＇ t always determine the clustering number correctly. In the process of iteration, when the FISODATA al- gorithm is introduced into the fuzzy set theory can accurately determine the clustering number.