摘要
协同多源遥感影像、融合多特征参数是目标地物精确识别的有效方法,但过多的特征参数会造成数据冗余,降低分类精度。该文以岩溶地貌甘蔗种植区为例,使用Sentinel-1,Sentinel-2影像数据以及SRTM数字高程数据提取研究区地物的光谱特征、指数特征、纹理特征、地形特征和极化特征,其中,指数特征考虑了众多遥感传感器少有的红边波段计算的红边指数,纹理特征加入了雷达影像纹理。在实验中设计了6种方案探讨不同影像特征以及基于随机森林优选的最佳特征组合对甘蔗提取的影响。结果表明:在光谱特征叠加不同特征类型对研究区地物进行分类的情况下,不同特征类型重要性排序为光谱特征>指数特征>纹理特征>地形特征>极化特征;6种方案中基于随机森林算法构建的优选特征方案融合了不同特征变量,其甘蔗提取效果最佳,用户精度和生产者精度都高于97%,总体精度为95.49%,Kappa系数为0.94。
The integration of multi-source remote sensing images and multi-feature parameters is effective in the accurate identification of target ground objects.However,excess feature parameters can cause data redundancy,reducing classification accuracy.Focusing on a sugarcane planting area with Karst landforms,this study extracted the spectral,index,texture,topographic,and polarization features of the ground objects in the study area from Sentinel-1/2 images and SRTM digital elevation data.The index features involved the red edge index calculated based on the red-edge band,which was scarce in data derived from remote sensing sensors,and the texture features included the Radar image textures.In the experiment,six schemes were designed to explore the effects of different image features and the random forest-based optimal feature association on sugarcane information extraction.The results show that for the classification of ground objects in the study area using spectral features combined with other feature types,the importance of the feature types ranked in descending order of spectral features,index features,texture features,topographic features,and polarization features.Among the six schemes,the scheme based on the random forest algorithm,integrating different feature variables,yielded the optimal information extraction effect for sugarcane,with both user and producer accuracy higher than 97%,overall accuracy of 95.49%,and a Kappa coefficient of 0.94.
作者
卢献健
张焕铃
晏红波
黎振宝
郭子扬
LU Xianjian;ZHANG Huanling;YAN Hongbo;LI Zhenbao;GUO Ziyang(College of Geomatics and Geoinformation,Guilin University of Technology,Guilin 541006,China)
出处
《自然资源遥感》
CSCD
北大核心
2024年第1期86-94,共9页
Remote Sensing for Natural Resources
基金
广西自然科学基金项目“基于高分影像的喀斯特地区土壤水分反演关键问题研究”(编号:2022GXNSFBA035639)
国家自然科学基金项目“地基和星载GNSS-R融合的花岗岩滑坡高时空分辨率土壤湿度反演研究”(编号:42064003)
广西空间信息与测绘重点实验室开放基金项目“广西地区农业干旱遥感监测及预警方法研究”(编号:桂科能19-050-11-23)共同资助。
关键词
多源遥感
精准识别
随机森林
特征优选
红边波段
极化特征
multi-source remote sensing
accurate identification
random forest
optimal feature selection
red-edge band
polarization feature
