基于改进标记控制分水岭方法的树冠分割  

Crown segmentation based on improved watershed marker-controlled method

作　　者：武晓康 王良松 WU Xiaokang;WANG Liangsong(Guangzhou Conghua District Planning and Surveying Geographic Information Center,Guangzhou,Guangdong 510000,China;College of Geomatics and Geoinformation,Guilin University of Technology,Guilin,Guangxi 541004,China)

机构地区：[1]广州市从化区规划和测绘地理信息中心,广东广州510000 [2]桂林理工大学测绘地理信息学院,广西桂林541004

出　　处：《测绘标准化》2024年第4期49-55,共7页Standardization of Surveying and Mapping

摘　　要：针对传统分水岭方法从点云数据提取单木易出现树冠过分割的问题,本文提出一种改进标记控制分水岭的树冠分割方法。首先,用无人机激光雷达(LiDAR)点云数据滤波生成数字高程模型(DEM)、数字表面模型(DSM)和冠层高度模型(CHM),利用形态学开闭运算对CHM进行重构、去噪;其次,利用面向对象的方法基于无人机多光谱数据预先提取树冠,基于CHM将预先提取的树冠作为标记,运用分水岭变换提取单木树冠;最后,验证该方法的精度。结果表明,本文方法进行树木分割正确率大于91%,过分割及欠分割误差低于6%,相较于改进前正确分割率提高了17.14%、过分割误差降低了11.43%;改进标记控制分水岭法提取树高的决定系数(R2)高于0.89,均方根差(RMSE)低于1.1,标准均方根误差检验值(nRMSE)低于8.7%,表明本文方法可以有效减少过分割现象。For the reasons that the traditional watershed method of extracting individual tree from point cloud data is easier to result in problems such as over-segmentation of tree crowns,this paper proposes an improved tree crown segmentation method for marker-controlled watersheds.Firstly,the point cloud data from UAV LiDAR are filtered to generate digital elevation model(DEM),digital surface model(DSM)and canopy height model(CHM),and CHM is reconstructed and denoised by using morphological open and close operations.Secondly,the canopy is pre-extracted from the UAV multispectral data by using an object-oriented approach,and the individual tree canopy is extracted by using watershed transformation based on the CHM using the pre-extracted canopy as a marker.Finally,the accuracy of the method is validated.The results show that the tree segmentation correct rate by the method proposed in this paper is more than 91%,and the over-segmentation and under-segmentation errors are less than 6%,17.14%higher in correct segmentation rate and 11.43%lower in over-segmentation error compared with those used before the improvement,moreover,the correlation coefficient(R2)of the improved markers to control the watershed extraction of the tree heights is larger than 0.89,the root mean square error(RMSE)is less than 1.1,and the standardized root mean square error test value(nRMSE)is less than 8.7%,indicating that the method proposed in this paper can effectively reduce the over-segmentation error.

关 键 词：无人机激光雷达 多光谱 树冠 单木分割 分水岭 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]