首页|基于空间异质运算的结构信息提取辅助遥感影像分类研究

基于空间异质运算的结构信息提取辅助遥感影像分类研究

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
以Landsat-8和高分一号数据为例,采用仅有光谱特征、3种纹理特征(概率统计、灰度共生矩阵、空间异质运算)辅助光谱特征的方法提取影像空-谱信息,并通过支持向量机分类器进行基于像元的地物分类.结果表明:1)纹理特征辅助光谱特征的地物分类精度明显优于仅使用光谱特征的分类,可提高8.62％～24.36％;2)相较于概率统计、灰度共生矩阵方法结果,空间异质运算结果分类精度在GF-1影像中分别提高了 13.31％和2.03％,在Landsat-8影像中分别提高了 11.62％和7.79％;3)对于线状地物,相较于概率统计、灰度共生矩阵方法结果,空间异质运算结果分类精度在GF-1数据中分别提高了 29.31％和0.80％,在Landsat-8数据中分别提高了 11.90％和6.64％,有效减小了分类误差.因此,空间异质运算提取的空间结构信息辅助光谱特征的分类方法能显著改善遥感图像的分类精度,为空间结构信息辅助遥感影像地物分类及线状地物的提取提供一种新的思路和方法.
Remote Sensing Image Classification Based on the Structural Information Extraction of Spatial Heterogeneity Operation
In order to solve the problems of texture extraction methods existed in remote sensing image clas-sification,such as narrow recognition range,difficult boundary distinction,low model applicability and low classification accuracy of linear ground objects,this paper explored how to make full use of high order tex-ture features reflected by spatial heterogeneity operation among remote sensing image elements to improve classification accuracy.Landsat-8 and GF-1 data were taken as examples,spectral feature sand three texture features(occurrence measures,gray co-occurrence matrix,spatial heterogeneity operation)combined with spectral features were used to extract spatial and spectral information of remote sensing images,then pixel-based support vector machine classifier was used to classify ground objects.The results indicated that 1)the classification accuracy of texture features combined with spectral features was significantly better than that of spectral features,and the overall accuracy could be improved by 8.62％-24.36％.2)Compared with the results of occurrence measures and gray co-occurrence matrix,the classification accuracies of spa-tial heterogeneity operation in GF-1 data improved by 13.31％and 2.03％respectively,and those in Land-sat-8 data improved by 11.62％and 7.79％respectively.3)For land cover types with linear features,com-pared with the results of occurrence measures and gray co-occurrence matrix,the average classification ac-curacies of spatial heterogeneity operation in GF-1 data increased by 29.31％and 0.80％respectively,and those in Landsat-8 data increased by 11.90％and 6.64％,respectively,effectively reducing classification er-rors.Therefore,the classification method of spatial heterogeneity operation to extract structural informa-tion combined with spectral features can obviously solve the current classification difficulties,and provide a new way for spatial structure information assisted remote sensing image land use classification and linear feature extraction.

linear objectspatial heterogeneity operationtextural featureground-object classificationre-mote sensing image

裴晨阳、张廷龙、高焕霖、张青峰

展开 >

西北农林科技大学资源环境学院,陕西杨陵 712100

线状地物 空间异质运算 纹理特征 地物分类 遥感影像

陕西省自然科学基础研究计划国家自然科学基金国家自然科学基金联盟软科学一陕西省农业土地空间规划及治理研究项目联盟软科学一陕西省农业土地空间规划及治理研究项目

2021JZ-174170123941301451F2021422001LMR202204

2024

10.3969/j.issn.1001-7461.2024.03.22

西北林学院学报
西北农林科技大学

西北林学院学报

CSTPCD北大核心
影响因子：1.078
ISSN：1001-7461
年,卷(期)：2024.39(3)
  • 46