基于区间二型模糊模型的高分辨率遥感影像分割方法

High resolution remote sensing image segmentation based on the interval type-2 fuzzy model

为了解决高分辨遥感影像分割过程中存在的类属和分割决策不确定性问题,提出一种基于区间二型模糊模型的监督分割方法。利用高斯函数建立影像分割一型模糊模型,其模糊隶属度表征像素类属不确定性;为了改善一型模糊模型隶属度对类属不确定性的表达以及增强分割决策的准确性,分别采用均值和方差扩展方法在上述一型模糊模型基础上构建区间二型模糊模型;综合利用一型模糊模型隶属度及二型模糊模型隶属区间等信息实现分割决策。提出算法和经典算法对合成影像、World View-II和IKONOS全色影像进行分割实验,定性和定量的分析结果表明提出算法具有更高的分割精度。

With the development of the technology of remote sensing data collection,there are various new types of remote sensing data with high spatial resolution. Image segmentation technology is a basis and important mission in remote sensing image processing. High resolution imagery contains more detailed information about the object which provides many potentials and advantages in precise image segmentation. However,high resolution increases the uncertainty of the classification of the pixel and brings new problems and difficulties occur for the segmentation decision. Therefore,conventional segmentation algorithms,which are suitable for low and middle resolution imagery,cannot satisfy the demands of high resolution image segmentation. To solve the above problems,a supervised image segmentation algorithm is presented with an interval Type-2 fuzzy model. The Gaussian distribution is used to establish a Type-1 fuzzy image segmentation model,where the fuzzy membership function describes the uncertainty of the classification of the pixel. To improve the representation of the fuzzy membership function in type-1 fuzzy model,its mean and variance are extended to establish an interval Type-2fuzzy model. The proposed algorithm adopts the Type-1 and type-2 Fuzzy models,which realizes the image segmentation decision. Experiments on the synthetic,World View-II and IKONOS panchromatic images are carried out,and the qualitative and quantitative analyses demonstrate that the proposed algorithm has a higher accuracy than others.