基于植被指数、光谱和形状指数的面向对象沙地提取方法研究

Extracting Sandy Lands Using an Object-oriented Method Based on Vegetation Indices,Spectral Mean and Shape Index

以辽宁省康平县为研究区,基于高分一号卫星(GF-1)空间分辨率16 m的多光谱影像,通过分析沙地提取的最优分割与合并尺度,以及沙地与其他地物的植被指数、光谱及形状指数特征差异,构建沙地特征提取规则。采用面向对象方法对沙地信息进行提取。结果表明:辽宁省康平县沙地提取的遥感影像最优分割尺度为40,最优合并尺度为90。沙地与非沙地的归一化植被指数和绿光波段光谱均具有高可分性,Jeffries-Matusita(J-M)距离均达到1.92以上。基于植被指数、光谱特征和形状指数3个参数的面向对象沙地提取结果总体精度达到88.09%以上,Kappa系数高于0.81。此方法具有较好的适用性,能够为沙地信息提取和面积估算提供参考。

Kangping county in Liaoning province,China was selected as a case in this study.Using multispectral images with 16-m spatial resolution from the GF-1 satellite,rules for extracting sandy lands boundaries were developed by analyzing the optimal segmentation and combination scales for sandy lands extraction.The differences between vegetation indices,spectral means and shape indices for sandy lands and non-sandy lands were also analyzed.The sandy lands information was extracted by using an object-oriented method.The results showed that the optimal segmentation scale was 40 and optimal combination scale was 90 in Kangping county.Both the NDVI and green-band spectral mean between sandy lands and non-sandy lands were highly separable,with a Jeffries-Matusita distance over 1.92.The overall accuracy of sandy lands extraction was over 88.09%using an object-oriented method which based on three parameters including vegetation indices,spectral mean and shape index.The Kappa coefficient was higher than 0.81.This method is relatively easy to apply and it provides a reference for extracting information on and estimating the area of sandy lands.