基于不同分辨率遥感影像自动提取切沟的精度分析和转换模型

Accuracy Analysis and Conversion Model of Gully Automatic Extraction based on Remote Sensing Images with Different Resolutions

黄土高原地貌类型独特而复杂,切沟侵蚀是塑造该区地貌的主要动力之一。研究不同分辨率遥感影像提取切沟的适用性和自动提取方法,可为切沟侵蚀遥感监测和沟蚀防治等提供有效手段。以黄土高原南部山西吉县残塬沟壑区为研究区,使用面向对象分析方法和随机森林分类算法分别从0.5 m Google影像、2 m GF-1融合影像和8 m GF-1多光谱影像中自动提取切沟,分析提取精度,并构建转换模型,提高低分辨率遥感影像提取的切沟沟长、面积参数的精度。结果表明:(1)依据特征类别,特征变量对于切沟识别的重要性排序如下:光谱特征>纹理特征>几何特征。(2)0.5 m和2 m分辨率影像切沟分类精度较高,生产者精度和用户精度均达90%以上,8 m GF-1影像切沟分类的生产者精度和用户精度为85%左右。(3)0.5 m和2 m分辨率影像提取的切沟沟长和沟宽的百分误差分别为5%和13%左右;8 m分辨率影像提取的切沟沟长、面积和沟宽的平均百分误差为18.82%、27.62%和18.93%。(4)基于0.5 m分辨率Google影像提取的切沟形态特征参数,建立8 m分辨率GF-1影像提取的切沟沟长转换模型(L=1.22L’-0.28)和面积转换模型(A=1.44A’+31.56),转换结果具有较高的精度。

Gully erosion is the major driver of land degradation and the unique landforms on the Loess Plateau.It is of practical significance to assess the applicability of extracting gully from satellite images with different resolutions and explore automatic gully extraction method.Google image(0.5 m resolution)and GF-1 images(2 m and 8 m resolution)were used to extract gullies automatically with object-based image analysis and random forest in Zhongduo tableland located in the southeastern Loess Plateau.Gully morphological parameters of30 gullies extracted from three satellite images were compared to those from UAV data(0.14 m resolution).The results were as follows:(1)The importance of image feature variables used for gully extraction is sorted as follows:spectral feature>texture feature>geometric feature.(2)The user accuracy and producer accuracy of gully extraction based on 0.5 m and 2 m resolution images were higher than 90%,while the user accuracy and producer accuracy reduced to 85%when 8 m resolution image was used.(3)The errors of gully length and width extracted from 0.5 m and 2 m resolution images were about 5%and 13%.The average error of extracted gully length,area and width from 8 m resolution image were 18.82%,27.62%and 18.93%,respectively.(4)A model was put forward for improving the accuracy of gully length and gully area extracted from GF-1 image with 8 m resolution,based on the gully parameters extracted from 0.5 m resolution image,i.e.,L=1.22L’-0.28,R^(2)=0.896 and A=1.44A’+31.56,R^(2)=0.916.