鹤地水库SPOT7影像分类研究

SPOT7 image classification of Hedi reservoir

针对水库地区开展的高分辨率影像地物信息提取研究少而需求大以及面向对象的影像分割尺度选择难问题,基于SPOT7高分辨率卫星影像数据,选取广东省湛江市鹤地水库作为研究区域,提出一种针对高分辨率影像的最优分割尺度选择以及针对水库地区的典型地物模糊分类规则构建方法,成功实现水体、植被、裸土、建成区的快速准确提取。结果表明:通过引入局部方差变量和Moran’I指数并结合最小风险贝叶斯决策原则能直观、准确、唯一地识别最佳空间分割尺度;通过分析光谱和纹理特征发现,水体和植被的光谱特征显著,建成区的纹理特征表现突出,而裸土单靠光谱特征不足以辨识,需要纹理特征的辅助;针对典型地物构建的模糊规则分类效果整体良好,总体精度达到了90.8%,Kappa系数为0.884。本研究有助于水库管理部门、国土资源调查部门快速、动态地把握库区的整体用地情况。

Aiming at the problem that fewer researches focus on the extracting ground objects from high-resolution image of reservoir area while great demand,and how to choose the optimal scale to conduct object-oriented image segmentation.This study proposed a new methods for the selection of optimal segmentation scale and constructing fuzzy rule sets for land cover classification of reservoir area based on SPOT-7 high resolution satellite images,taking Hedi Reservoir as study area,succeeded in extracting the typical land cover types like water,vegetation,soil and built-up.The results indicate that:by introducing local variance and Moran’I index,based on the principle of minimum risk Bayesian decision,the optimal segmentation scale can be identified intuitively,accurately and uniquely;by analyzing the spectral and texture features,it is easy to find that water and vegetation show significant spectral characteristic,built-up shows obvious texture characteristic,however,soil can’t be distinguished only by spectral features,which needs the assistance of texture.The fuzzy rule classification method achieves a good overall performance for typical land cover types,all land cover information are extracted completely,with an overall accuracy of 90.8% and the Kappa coefficient is 0.884.This study helps reservoir management department and land and resources survey department to quickly and dynamically grasp the whole land information of reservoirs.