基于区间二型模糊模型的高分辨率遥感影像分割方法研究

High spatial resolution remote sensing image contains detail information of the Earth surface. It brings benefit for further development of the detected object while makes the distributions and connections both inside and between different objects more complex. And that greatly increases the uncertainties of the subordination of pixels and modeling. So it is an important task to characterize the uncertainties mentioned above and model the distributions of homogeneous regions, effectively. Aiming on the point, the project takes a further research on more High spatial resolution remote sensing image contains detail information of the Earth surface. It brings benefit for further development of the detected object while makes the distributions and connections both inside and between different objects more complex. And that greatly increases the uncertainties of the subordination of pixels and models. So it is an important task to characterize the uncertainties mentioned above and model the distributions of homogeneous regions, effectively. Aiming on the point, the project takes a further research on more accuracy segmentation for high resolution remote sensing images and proposes a supervised segmentation method, which can be used on high spatial resolution remote sensing images, based on the interval type II fuzzy model. Employing the interval type II fuzzy theory, fuzzy clustering and neural networks, the research of this project will focus on (1) constructing type-I fuzzy membership function model to express the uncertainty of pixels, (2) exploring the effect of the interval type-II fuzzy model and the range of its uncertainty on the accuracy of segmentation, (3) proposing the decision model for type-I fuzzy model and interval type-II fuzzy model, (4) Describing the relationship between neighbor pixels. This project proposes a type-II fuzzy model theory which can be generally used to high resolution remote sensing images segmentation and provides a new method for high accuracy remote sensing image segmentation.

高空间分辨率遥感影像地表覆盖及细节信息丰富，地物目标复杂多样，像素光谱测度的空间相关性及同一地物目标内像素光谱测度分布特征复杂，使像素类属的不确定性及建模的不确定性增大。因此，建立准确可靠的不确定特征模型及分割决策模型是实现高分辨率遥感影像精准分割的关键。针对以上关键因素，对高分辨率遥感影像精准分割算法展开深入的理论与实践研究，提出适于高分辨遥感影像的基于区间二型模糊模型的监督分割方法。以二型模糊理论、模糊聚类、神经网络等理论为基础，项目重点研究(1)构建表达像素类属不确定性的一型模糊隶属函数模型；(2)构建表达模型不确定性的区间二型模糊隶属函数模型及不确定区域范围对地物分割的影响规律研究；(3)构建基于一型和二型模糊模型的分割决策模型；(4)刻画邻域像素空间关系。项目研究成果将给出普适的高分辨率遥感影像二型模糊建模方法学，为高分辨率遥感影像的精准解解译提供一种新思路。

高空间分辨率遥感影像地表覆盖及细节信息丰富，地物目标复杂多样，像素光谱测度的空 间相关性及同一地物目标内像素光谱测度分布特征复杂，使像素类属的不确定性及建模的不确 定性增大。因此，建立准确可靠的不确定特征模型及分割决策模型是实现高分辨率遥感影像精准分割的关键。针对以上关键因素，对高分辨率遥感影像精准分割算法展开深入的理论与实践研究，提出适于高分辨遥感影像的基于区间二型模糊模型的监督分割方法。.以二型模糊理论、模糊聚类、积极学习等理论为基础，项目主要研究了以下三方面内容,.(1)构建表达像素类属不确定性和空间相关性的一型高斯回归模型；.(2)构建表达模型不确定性的区间二型模糊隶属函数模型及不确定区域范围并对其进行三维可视化表达，实现区间二型高斯回归模型内在机理研究；.(3)构建基于一型和二型模糊模型的分类决策模型.得到结果如下：.(1)给出一种具有一般性及可扩展性的基于区间二型模糊理论的影像建模框架，为高分辨率遥感影像的精准解解译提供一种新思路，助力二型模糊理论应用与发展 .(2)对项目中提出的区间二型高斯归回模型内在机理进行初步的探索，发展和完善了区间二型模糊理论。.其中的关键数据是:.(1)实现了的区间二型模糊模型不确定区域的自适应确定.(2)实现了区间二型高斯回归模型多维隶属度数据的三维可视化表达，以最直观方式的呈现了模型本身特征、地物间相关特性，二型模糊划分的内在机理。.研究成果可为智慧农业，智慧城市及智慧交通等领域精准快速地提供不可或缺的地物及目标的属性特征；可为基于遥感影像的大规模土地覆盖及利用、地物目标辨识及其特征提取等应用提供有效手段，促进高分辨率遥感影像在国土资源调查、灾害预测、环境监测等领域的广泛应用。作为一种新技术和新方法，目前该领域研究和应用还处于初始阶段，鲜有研究者对此开展系统性研究工作。因此，项目也试图填补这方面研究的空白。