基于LBP纹理特征的赣南柑橘果园提取  

Gannan citrus orchard extraction based on LBP texture features

作　　者：陈俊杰 CHEN Junjie(of Civil Engineering and Surveying and Mapping,Jiangxi University of Technology,342300,Ganzhou,Jiangxi)

机构地区：[1]江西理工大学土木与测绘工程学院,江西赣州341000

出　　处：《长江信息通信》2023年第8期47-50,共4页Changjiang Information & Communications

摘　　要：文章采用GF-7号和GF-1号遥感影像作为数据源,构建不同空间尺度序列的研究区影像,通过LBP算子方法提取纹理特征,然后将不同空间分辨率影像中提取的LBP纹理特征与光谱特征、植被指数特征组合构建特征集,使用随机森林分类方法对研究区进行分类和柑橘果园识别提取,最终发现加入LBP纹理特征信息可以显著降低柑橘果园的错分误差和漏分误差,有效提升柑橘果园提取精度和总体分类精度。0.65m~8m空间分辨率影像中,2m分辨率下结合LBP算子的提取精度最高,总体精度和柑橘果园提取精度分别达到94.87%和93.24%。In order to obtain the distribution information of citrus orchards in Gannan region from high-resolution images more quickly and accurately,this paper uses GF-7 remote sensing images with high spatial resolution as the data source,extracts tex-ture features by GLCM,LBP operator and Laws texture energy measurement,and then combines the texture features extracted by different methods with spectral features and vegetation index features to construct four different feature combination sche-mes.Then,the classification experiment of these four schemes is carried out by the random forest classification algorithm,and the experimental results of different texture features added to the classification scheme are compared.The results show that the classification effect after adding texture features is significantly better than that of only combining spectral features and vegeta-tion index features.The effect of identifying Gannan citrus orchards combined with LBP operator texture features was better than GLCM and Laws,and the extraction accuracy(Fa)and overall classification accuracy(OA)of Gannan citrus orchards re-ached 92.89%and 94.34%.

关 键 词：高分辨率遥感 LBP算子 随机森林分类 柑橘果园提取 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]