黄土丘陵沟壑区切沟分布的遥感图像识别最优尺度研究

Research for Optimal Scale on the Remote Sensing Image Recognition of Gully Distribution in Loess Hilly-Gully Region

本研究提出了一种基于最优尺度分割和阈值统计优化的切沟提取新方法。首先,通过灰度共生矩阵(GLCM)内部对象方差和空间自相关系数,建立了一个最优分割尺度选择目标函数,经试验得到了研究区内农路、切沟、耕地(山坡地、梯田)、林地的最优分割尺度分别为24,31,36,42;在影像分割基础上,提出了切沟提取阈值统计优化方法,即结合NDVI、光谱亮度值(BRIGHTNESS)和纹理(GLCM)特征变量,通过K-means和两步聚类法(TwoStep Cluster),分别确定切沟候选对象和假阳性对象的不同分类阈值,为切沟提取打下基础;经过t-statistics检验,显示光谱亮度值和纹理变量在95%的置信水平上均显著,从而表明提取阈值具有稳定性和可靠性。

This paper brought forward a new method of optimal segmentation scale and threshold methods.maximum of objective function was established by using Moran′s I index and intrasegment variance of GLCM and was tested in study area,the results of the optimal partition scale were 24,31,36,42 for agriculture road,gully,cultivated land(slope,terrace)and the forestland in research area.Based on optimal segmentation scale selection,the threshold methods estimated by K-means cluster analysis and TwoStep Cluster using NDVI,BRIGHTNESS and GLCM were established to identify gully candidates and separate gullies from false positives.In order to find out whether the clusters are sufficiently different to represent the false positives,a t-test performed for GLCM,BRIGHTNESS showed that those tome parameters were significant at95%of the confidence level,therefoe,suggested that the threshold of clusters were reliable and robust.