Abstract: Taking Honghu wetland in Hubei province as the study area, and using two ETM+ images acquired on July 27, 2002 and July 9, 2005 respectively, a new wetland change detection method based on improved Independent Component Analysis (ICA) was poposed. Firstly, the traditional ICA algorithm was improved in two aspects: the initial value determination and iterative process. Secondly, Tasseled cap transformation was taken on original images and their wetland components were extracted for ICA transformation. Finally, based on the obtained independent component that represents the possible changed areas, adaptive threshold segmentation method was used to get the final changed areas. The experimental results show that the proposed algorithm has higher accuracy than traditional ICA algorithm and post-classification comparison algorithm.

Key words: wetland, tasseled cap transformation, Independent Component Analysis (ICA), change detection

摘要： 以湖北省洪湖湿地为研究区，采用2000年7月27日和2002年7月9日ETM+多光谱影像作为遥感信息源，提出了一种基于改进独立分量分析（ICA）的湿地变化检测方法。首先对传统ICA从初始值选取和迭代过程两方面加以了改进。然后对原始多波段影像进行缨帽变换，提取前后时相的湿度分量作为输入的混合影像，进行改进ICA变换。最后在变换得到的独立分量基础上，对代表变化区域的独立分量进行自适应阈值分割，得到湿地变化检测结果。实验表明，改进ICA法检测精度较传统ICA法及分类后比较法有所提高。

关键词: 湿地, 缨帽变换, 独立分量分析, 变化检测