基于傅里叶描述子的高分辨率遥感图像地物形状特征表达

Shape feature representation of ground objects from high-resolution remotely sensed imagery base on Fourier Descriptors

本文在对傅里叶描述子进行归一化的基础上,将该方法引入地物轮廓的形状特征描述中,针对建筑物、农田、道路和河道4种典型地物,分别从谱线特征、不同频段描述子对形状特征的贡献率、形状重构三个方面进行分析,结果表明,在谱线图中,直流分量对形状特征的贡献率在70%以上,低频和高频成分共占7%—24%左右,中频成分的贡献率只有2%—4%左右,仅低频成分(第1—5项)便能够很好地进行地物形状重构。最后将第1—5项描述子应用到基于决策树的面向对象分类中,得出实验区总体分类精度为98.48%,Kappa系数为0.9714。傅里叶描述子的方法能够很好的表达高分辨率遥感图像的地物形状特征。

The traditional Fourier Descriptors（FDs） are normalized in this paper to make it independent of translation, rotation and scale changes.Four typical objects i.e.building,paddy,road and river are selected and their boundaries are expressed as sequences of complex numbers.FDs are obtained through one-dimensional Fourier transform.The characteristics of the frequency spectrum,contribution rate and the shape reconstruction are analyzed.The results show that the different frequency ranges have different contribution rates;the Direct Component（DC） reaches a proportion of more than 70%;the Low Frequency（LF） and High Frequency（HF） totally reach 7%-24%while the Medium Frequency（MF） merely 2%-4%.The LF components（descriptors 1—5） make a commendable reconstruction of objects＇ shape and these descriptors are applied to the object-oriented classification.The overall classification accuracy is 98.48%with a Kappa coefficient 0.9714.