文献类型：期刊文献

中文题名：基于面向对象多特征变化向量分析法的森林资源变化检测

英文题名：Object-oriented Forest Change Detection Based on Multi-feature Change Vector Analysis

作者：王晓慧[1] 谭炳香[1] 李世明[1] 冯林艳[1]

第一作者：王晓慧

机构：[1]中国林业科学研究院资源信息研究所,国家林业和草原局林业遥感与信息技术重点实验室,北京100091

年份：2021

卷号：34

期号：1

起止页码：98-105

中文期刊名：林业科学研究

外文期刊名：Forest Research

收录：CSTPCD;;Scopus;北大核心:【北大核心2020】；CSCD:【CSCD2021_2022】；

基金：中央级公益性科研院所基本科研业务费专项资金(CAFYBB2017MB012);国家重点研发计划课题(2017YFC0506502);中国林科院资源所项目(2019JYZJ05)。

语种：中文

中文关键词：面向对象;变化向量分析法;高分二号多光谱影像;特征选择;森林资源变化检测

外文关键词：object-oriented;change vector analysis;GF2 multispectral image;feature selection;forest change detection

分类号：S771.8

摘要：[目的]基于面向对象变化向量分析法,进行森林资源变化检测。[方法]应用国产高分二号多光谱影像,以森林采伐和造林活动多、林地变化频率高的广西壮族自治区上思县为研究区,应用随机森林平均精确率减少的方法进行变化特征的选择,通过选取的不同特征向量和常规的基于光谱均值、光谱均值和标准差的变化向量分析法,以及基于NDVI差值法的变化检测结果对比,获取较好的森林资源变化检测方法和结果。[结果]高分二号多光谱影像的蓝、绿、红波段光谱均值和NDVI值共4个特征参与的变化向量分析法,识别森林资源变化精度高,总体精度92.94%,Kappa系数0.7630,变化地类误检率15.63%,漏检率22.86%。[结论]经过特征选择后,基于面向对象变化向量分析法比常规的多特征参与的变化向量分析法识别森林资源变化的效果好。
[Objective]To detect forest resources with object-oriented multi-feature change vector analysis.[Method]Object-oriented change vector analysis was applied to detect forest changes using GF2 multispectral images in a study site located in Shangsi County,Guangxi Zhuang Autonomous Region where forest harvesting and planting often happens and the forest land experiences frequent changes.Random forests with mean decrease accuracy were employed to select features.Better method and result of forest change detection were produced by comparison between change vector analysis by feature selection and general change vector analysis based on spectral average,spectral average and standard deviation,as well as NDVI difference.[Result]The accuracy of forest change detection based on change vector analysis with NDVI and spectral averages of blue,green and red bands was the highest.The overall accuracy was 92.94%,the Kappa coefficient was 0.7630,the commission rate and omission rate of change land type were 15.63%,and 22.86%respectively.[Conclusion]By feature selection,object-oriented change vector analysis shows better effect on forest change detection than general change vector analysis.