面向对象的湿地景观遥感分类——以杭州湾南岸地区为例

在ENVIEX软件的Feature Extraction平台上,利用LandsatTM影像数据,采用面向对象方法对杭州湾南岸地区湿地景观进行遥感影像分类;通过与基于最大似然法、人工神经网络法、支持向量机法等传统像元方法的相应分类结果进行比较,系统分析了面向对象方法在中低分辨率遥感影像的湿地景观生态分类中的有效性。研究结果表明:①较之单一依据像元光谱值进行分类的传统方法,面向对象方法综合考虑了对象的光谱、空间、纹理、色彩等多种属性特征,因而对于类型复杂多样、分布界限模糊、光谱混淆与混合像元现象严重的沿海滩涂、湖泊、河流等湿地景观具有更好的鉴别能力,也因此获得更高的分类精度(研究区景观分类总精度为88.80%,Kappa系数为0.8765);②面向对象方法在分类中提取的是由同质性像元组成的"对象",且在合理的影像分割下得到的对象破碎化程度较低,因而能在较大程度上减小分类结果中的"椒盐噪声"干扰;而基于像元方法提取的景观类型以离散像元形式组成,难以清晰表征景观的边界、形状等信息,所以分类结果中会有明显的噪声现象;③影像分割在运用面向对象方法进行遥感影像分类过程中具有重要影响,实验结果表明,60%的分割尺度和归并尺度组合较有利于中低分辨率影像的遥感分类;④面向对象分类过程中诸如影像分割精度的评价、最优分割尺度的选取、特征空间的优化等问题,则有待今后进一步探讨。

Object-oriented Classification for Satellite Remote Sensing of Wetlands:A Case Study in Southern Hangzhou Bay Area

Based on data of Landsat TM imagery and the Feature Extraction module of ENVI EX software,object-oriented method was applied to the remote sensing image classification of wetlands in southern Hangzhou Bay area.And analytical comparisons of object-oriented classification and traditional pixel-based methods(maximum likelihood method,artificial neural net method and support vector machines method) were conducted on the basis of the classification results in order to systematically assess the utility of object-oriented classification of wetlands with medium and low resolution satellite imagery.The research showed that compared with the traditional classification methods which only consider the spectral characteristics of the targets,object-oriented classification comprehensively utilizes more detailed information of the remote sensing imagery(e.g.,spectral characteristics,texture feature,spatial relationship,color space,band ration).Thus,it gains a better discriminability for some specific wetland landscapes(e.g.,intertidal mud,lake wetlands,river wetlands),which involve complex forms,fuzzy boundaries,and severe spectral and pixel mixed problems,and yields higher classification accuracy(an overall accuracy of 88.80% and a Kappa coefficient of 0.876 5 were achieved in this study).Object-oriented method extracts the so called ’object’ which consists of some homogeneous pixels in the process of classification,and the objects show a low degree of fragmentation through a reasonable level of image segmentation.Therefore,this method significantly reduces the disturbance of salt-and-pepper noise in the classification results and generates more accurate results.In contrast,the landscapes extracted by pixel-based methods are made up of discrete pixels,fall short of representing the shapes and boundaries of the targets clearly,and also fail to block out the noise phenomena.It is found that image segmentation plays an essential role in the process of object-oriented classification,and 60% is a proper parameter of segmentation and merger scale for the object-oriented classification of medium and low resolution satellite imagery.Furthermore,the Feature Extraction module of ENVI EX enables a preview of the results of image segmentation real-time as a good approach for image segmentation,one that is more convenient and effective than other comparable softwares.Some important scientific issues existing in the process of object-oriented classification,e.g.,the evaluation of image segmentation precision,the identification of optimal segmentation scale,the optimization of feature space,entail further research.