摘要
【目的】及时、准确地掌握林地信息是森林经营管理的前提,高分辨率遥感影像为林地信息精细识别提供了可能。【方法】以当阳市玉泉乡为研究区,以国产卫星高景一号(SV-1)遥感影像为数据源,提取各波段光谱信息和植被指数作为分类特征,采用特征可分性、重要性及特征间冗余度分别构建了4种特征评价准则,基于支持向量机(SVM)分类器对研究区进行林地信息提取,结合森林资源二类调查结果进行精度验证。【结果】1、评价准则中,特征重要性优于可分性,特征可分性受高度相关的特征组合(如OSAVI和NDVI等)的影响会造成分类精度的下降。2、在特征重要性和可分性的基础上结合特征间冗余度能进一步提高分类精度并有效降低特征维数,特征维数由11维降至8维,特征可分性方法和特征重要性的分类精度分别提高了4.65%和4.58%;3、根据特征重要性结合冗余度选择RGVI、EVI、B1、B3、B2、DVI、RVI、Brightness 8个特征,建立SVM线性核分类模型可以达到最优分类效果,总体分类精度高达92.49%,Kappa系数为0.9084。【结论】SV-1遥感影像由于其高空间分辨率在林地信息精细提取中具有可行性,本研究通过建立特征评价准则筛选分类特征能进一步挖掘分类器的泛化能力,为及时、准确地获取林地信息提供技术支撑,同时也为同等高分辨率遥感卫星数据处理提供了参考。
【Objective】Timely and accurate mastery of forest land information is the premise of forest resource management.High spatialresolution remote sensing images provide the possibility of fine identification of forest land information.【Method】Taking Yuquan of Dangyang county as the study area,the Super View-1 image in October 2018 as the data source,single-band spectral information and vegetation index were extracted as classification features.Four evaluation criteria were constructed based on class separability,feature importance and feature redundancy.Then two classification models of linear kernel and Gauss kernel function of support vector machine were established to extract forest information in the study area.The overall accuracy was verified with the results of forest resource investigation.【Result】1、In the evaluation criteria,feature class separability,which is affected by highly correlated feature combinations(OSAVI and NDVI),is slightly inferior to feature importance.2、On the basis of feature importance and class separability,considering the redundancy between features can effectively reduce the feature dimension and further improve the classification accuracy.The feature dimension is reduced from 11 to 8,and the classification accuracy is improved by 4.65%and 4.58%respectively.3、According to the evaluation criteria,eight features of RGVI,EVI,B1,B3,B2,DVI,RVI and Brightness are selected.The optimal classification effect can be achieved by establishing the SVM linear kernel classification model.The overall classification accuracy is up to 92.49%,and the Kappa coefficient is 0.9084.【Conclusion】Super View-1 remote sensing image is feasible in detailed extracting forest land information because of its high spatial resolution,and the classification accuracy can be further improved by establishing feature evaluation criteria to select features,which will provide technical support for timely and accurate acquisition of forest land information.At the same time,it also provides reference for the
作者
曾文
林辉
李新宇
肖越
鲁宏旺
ZENG Wen;LIN Hui;LI Xinyu;XIAO Yue;LU Hongwang(Research Center of Forestry Remote Sensing&Information Engineering,Changsha 410004,Hunan,China;School of Forestry,Central South University Forestry&Technology,Changsha 410004,Hunan,China;Key Laboratory of Forestry Remote Sensing Based Big Data&Ecological Security for Hunan Province,Changsha 410004,Hunan,China;Key Laboratory of State Forestry and Grassland Administration on Forest Resources Management and Monitoring in Southern Area,Changsha 410004,Hunan,China)
出处
《中南林业科技大学学报》
CAS
CSCD
北大核心
2020年第7期32-40,共9页
Journal of Central South University of Forestry & Technology
基金
国家自然科学基金项目(31370639)。
关键词
林地信息提取
特征评价准则
支持向量机分类
高景一号
forest information extraction
feature evaluation criteria
support vector machine(SVM)classification
Super View-1
