变窗口纹理在SPOT5影像土地覆盖分类中的应用研究  

An Applied Research of Changing Window Texture in Land Cover Classification Using SPOT5 Image

作　　者：刘锦绣[1,2] 刘慧平[1,2] 吕颖[1,2] 穆晓东[1,2] 薛晓娟[1,2] 

机构地区：[1]北京师范大学地理学与遥感科学学院,北京100875 [2]北京师范大学遥感科学国家重点实验室,北京100875

出　　处：《地理与地理信息科学》2013年第5期26-30,F0002,共6页Geography and Geo-Information Science

基　　金：中央高校基本科研业务费专项资金资助项目;国家自然科学基金项目(40671127)

摘　　要：纹理作为有效的空间信息,在遥感信息提取中有着广泛的应用,其中纹理窗口的选择对分类精度影响较大,而且遥感影像中不同地物类型具有多尺度特性,利用单一窗口纹理信息进行分类不能使得各地物类型同时达到较高精度。该文采用SPOT5影像对研究区进行长序列窗口纹理分类实验,并提出变窗口纹理分类方法,实验区分类实验及精度对比表明:长序列窗口纹理分类实验中,7×7窗口分类精度最高,但在这一窗口下建设用地、耕地、裸地等地类的精度仍低于总体精度;而采用基于长序列窗口纹理分类后验概率的变窗口纹理分类方法,总精度达到86.17%,Kappa系数达到0.8230,与7×7窗口分类精度相比各地物类型的生产者精度与用户精度都有进一步提高,验证了变窗口纹理分类方法的可行性。Texture is regarded as effective spatial information and is widely used in the field of remote sensing information ex- traction. The selection of texture window size has great influence on classification accuracy, and different land cover types in re- mote sensing images have various scale characteristics, so using single window texture information to classify cannot achieve a high accuracy result for all land cover types. In this paper, SPOT5 image is used to carry out texture classification experiment u- sing long sequence of window and changing window texture classification experiment. The results and accuracy assessments in- dicate that using window size of 7 × 7 pixels can obtain the highest accuracy in the range of long sequence of window. However, the accuracy of built-up area, farm land and bare land are still below the overall accuracy. The changing window texture classifi- cation method based on posterior probability of long sequence of window can improve the accuracy further, and the overall accu- racy is 86.17%, the Kappa is 0. 8230. Comapred with the result of 7× 7 window size, the changing window texture classification result have a further increase in producer accuracy and user accuracy of different land cover types, testing the feasibility of chan- ging window texture classification method as well.

关 键 词：SPOT5影像 单窗口纹理 变窗口纹理 土地覆盖分类 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]