Abstract: In traditional FISODATA algorithm, the splitting and merging operations need to choose a threshold manly. Sometimes, unsuitable threshold can cause an incorrect number of classes. This paper presents an improved fuzzy ISODATA algorithm. In the proposed algorithm, a new operation is designed for splitting and merging based on fuzzy set theory and changes the threshold by the results of the segmentation after it splits or merges, which makes it “adaptive”. It solves the problems caused by choosing threshold manly. Experimental results from a series of synthetic and IKONOS image segmentation demonstrate the adaptability and accuracy of the proposed approach over competing with other methods.

Key words: remote sensing image segmentation, fuzzy clustering, fuzzy Iterative Self-Organizing Data Analysis Techniques Algorithm（ISODATA）

摘要： 传统模糊ISODATA（Fuzzy ISODATA，FISODATA）算法中，分裂-合并操作需人工选取阈值参数。而不适当的阈值往往使算法陷入局部极值，因而得到错误的类属数并最终影响图像分割结果。为此，在模糊集理论基础上提出一种改进的自适应FISODATA算法。该算法设计了自适应分裂-合并操作，即在每次分裂-合并后，根据该次计算结果改变参数阈值，解决了人为选取参数带来的诸多问题。利用该算法对模拟图像和真实IKONOS图像进行分割实验，均能得到良好的分割结果。

关键词: 遥感图像分割, 模糊聚类, 模糊迭代自组织数据分析技术算法（ISODATA）