基于自适应区间二型模糊聚类的遥感土地覆盖自动分类

遥感影像土地覆盖分类面临"类别密度差异显著"、"同谱异物"和"同物异谱"等不确定性问题,传统的分类方法(如FCM)因不能描述高阶模糊不确定性,无法完成准确建模,使分类误差较大,而二型模糊集恰是处理此类不确定性的有效工具.在引入二型模糊集新概念和自适应降型新方法的基础上,提出一种自适应二型模糊分类方法(A-IT2FCM):(1)基于样本集模糊距离度量构建面向分类的区间二型模糊集,以尽可能降低对先验知识和预设参数的依赖,从而满足自动分类的要求;(2)给出一种自适应探求等价一型代表(模糊)集合的高效降型方法,在此基础上进行自适应区间二型模糊聚类.实验数据为珠海横琴和北京颐和园的SPOT5影像数据,对比方法有AIT2FCM、基于Karnik-Mendel算法降型和基于Tizhoosh提出的简易降型方法的区间二型模糊C均值聚类以及作者前期研究提出的区间值模糊C-均值算法(IV-FCM).实验结果表明,A-IT2FCM方法分类效果佳,在类别具有较大密度差异和多重模糊性时能得到比FCM及IV-FCM更精确的边界和更连贯的类别,适于处理遥感影像土地覆盖类别的深层不确定性;同时在"光谱混叠"现象严重时,可以获得比对比方法更稳健、精度更高的影像自动分类结果,且时间复杂度明显低于基于Karnik-Mendel方法.

Land cover classification based on adaptive interval type-2 fuzzy clustering

There is great fuzzy uncertainty in the land cover classification using mid or high resolution remote sensing imagery, for example, different objects with the same spectra characteristics or the same object with different spectrums. The classic methods, such as FCM, are disable to carry out accurate modeling for the high-level fuzzy uncertainty, and then cause the classification error that should not be ignored in the application. However, the type-2 fuzzy sets is the tool to handle this type of uncertainty. An adaptive interval-valued type-2 fuzzy C-Means clustering algorithm (A-IT2FCM) is proposed based on the new ideas of the type-2 fuzzy sets and type reduction, including: (1) a new modeling method for interval-valued type-2 fuzzy set, which is on the basis of the fuzzy distance metric to reduce the dependency on the priori knowledge or default parameters as much as possible and meets requirements of auto-classification; (2) an effective type reduction approach by searching the equivalent type-1 fuzzy sets for the type-2 adaptively. The experimental data are three data windows of SPOT5 imagery from Zhuhai and Beijing, China. There are four different type-2 fuzzy clustering algorithms used for the auto land cover classification in this article: the algorithm based on Karnik-Mendel type reduction, interval-valued fuzzy C-Means clustering based on simple type reduction proposed by Tizhoosh, interval-valued fuzzy C-Means clustering proposed in our former study and A-IT2FCM presented in this article. The experimental results show that A-IT2FCM outperforms the compared algorithms. Especially when there is obvious density difference between objects and multiple fuzzy uncertainties in the experimental data, A-IT2FCM can achieve more accurate class boundaries and more coherent categories, which demonstrate that A-IT2FCM is suitable to process the deeper uncertainty in the remote sensing land cover classification. What is more, the computation complexity with A-IT2FCM is lower than that with Karnik-Mendel type reduction.