摘要
利用大丰市沿海滩涂湿地区域的高光谱影像和同时期的机载LIDAR数据,结合影像的光谱信息,采用随机森林算法(RF)对研究区进行湿地植被精细分类,并分析和评价分类模型参数设置对总体精度的影响,最后与SVM分类结果进行对比。结果表明:随机森林分类方法的总体精度为90.3%、卡帕(Kappa)系数为0.874;与传统的SVM分类方法相比,RF法均提高了4种湿地植被的生产者精度和使用者精度。通过分析RF分类模型参数设置对总体精度的影响,得出当生长树个数为30、生长树深度为30时,分类精度最高。
Based on the hyperspectral images and the same time airborne LIDAR data of Dafeng coastal wetland area,combined with the spectral information of the images,the random forest algorithm（ RF） was used to classify the wetland vegetation in the study area,and the influence of the parameters of the classification model on the overall accuracy was analyzed and evaluated. Finally,the results were compared with the SVM classification results. The results showed that the overall accuracy of the random forest classification method was 90.3% and the Kappa coefficient was 0.874. Compared with the traditional SVM classification method,the RF method improved the precision and user precision of the four wetland vegetation. By analyzing the influence of the parameters of the RF classification model on the overall accuracy,it is concluded that when the number of growing trees is 30 and the growth tree depth is 30,the classification accuracy is the highest.
作者
崔小芳
刘正军
CUI Xiaofang;LIU Zhengjun(Liaoning University of Engineering and Technology,Fuxin 123000,China;Institute of Photogrammetry and Remote Sensing,Chinese Academy of Surveying and Mapping,Beijing 100830,China;Henan Fenghua Engineering Technology Co.,Ltd.,Zhengzhou 450000,China)
出处
《测绘与空间地理信息》
2018年第8期113-116,共4页
Geomatics & Spatial Information Technology
基金
国家自然科学基金重点项目(41330750)资助
关键词
随机森林
多源遥感数据
湿地植被
精细分类
random forest
multi-source remote sensing data
wetland vegetation
fine classification
