摘要
根据城市植被在IKONOS影像上的光谱、纹理、几何和位置响应特征,采用面向对象的方法对城市植被进行分类.首先,利用NDVI和蓝波段光谱响应值的阈值将实验区分割为植被和非植被区,然后针对植被区利用区域增长算法进行二级分割,生成植被对象;根据植被在IKONOS上的响应特征,选择形状指数、亮度值、绿波段的最大差值、红波段的平均值、近红外波段的比率、近红外波段的方差和对象重心的位置,即横纵坐标以及Homogeneity指数等9个指标构建特征空间;在此基础上,利用最大似然法识别城市植被类型,并利用专家知识对分类结果进行再组合.研究表明,利用这种方法获得的城市植被信息总精度达到87.37%,Kappa系数达到0.8267.
Based on urban vegetation characteristics of spectrum,texture,geometry and location,the object-oriented method was used to identify urban vegetation categories from IKONOS imagery.Firstly,vegetated area and non-vegetated area were separated by the threshold of NDVI and digital number（DN）value of blue band,and further,region growing method was adopted to segment the vegetated area into ＂objects＂,which represent the aggregation of pixels adjacent in location and in DN value.Secondly,9 features of form index,bright,maximum spectral difference of green band,mean spectral value of red band,ratio of near-infrared band,the standard deviation of near-infrared band,x,y and Homogeneity,were selected to construct feature space.Thirdly,objects were classified into five vegetation categories using maximum likelihood methods and expert knowledge,within the feature space.The overall accuracy of 87.37% and the Kappa coefficient of 0.8267 of the classification results showed the object-oriented method is useful to classify urban vegetation.
出处
《浙江大学学报（农业与生命科学版）》
CAS
CSCD
北大核心
2007年第5期568-573,共6页
Journal of Zhejiang University:Agriculture and Life Sciences
基金
江苏省自然科学基金资助项目(BK2002420)
浙江省科技计划重点资助项目(2007C23089).
关键词
面向对象
城市植被
信息提取
object-oriented method
urban vegetation
information extraction
