融合机载LiDAR和植被指数的自适应单木提取方法  

Adaptive single tree extraction method based on fusion of airborne lidar and vegetation index

作　　者：代震 何荣[1] 王宏涛[1] 白伟森 DAI Zhen;HE Rong;WANG Hongtao;BAI Weisen(School of Surverying Land and Information Engineering,Henan Polytechnic University,Jiaozuo 454000,China)

机构地区：[1]河南理工大学测绘与国土信息工程学院,河南焦作454000

出　　处：《光学精密工程》2023年第22期3331-3344,共14页Optics and Precision Engineering

基　　金：NSFC-区域创新发展联合基金重点项目(No.U22A20566);河南省高等学校重点科研资助项目(No.18B420003);河南理工大学基本科研业务费专项资助项目(No.NSFRF170909)。

摘　　要：机载激光数据(Light Detection And Ranging,LiDAR)难以区分地面和草地范围,可见光植被指数无法分离灌木和乔木层,针对上述问题,构建一种融合LiDAR点云与RGB植被指数的多波段信息图像。以激光点云生成精细冠层高度模型(Canopy Height Model,CHM),利用无人机影像数据生成高清数字正射影像,在比较不同植被指数精度后选择差异增强植被指数(Differential Enhanced Vegetation Index,DEVI),与CHM进行融合。形态学重建CHM+DEVI融合图像,去除不合理值;构建训练样本,采用分类回归树算法,分割地面范围并自适应提取植被为乔木、灌木和草地,乔木区域采用局部最大值算法探测树顶点,作为前景标记,非乔木区域赋为后景标记,进行分水岭变换得到分割结果。将该方法提取的植被信息与实测数据进行精度分析,结果表明:改进方法在4个样方中,总体查全率提高3.2%,查准率提高3.9%,准确度F1得分提高3.5%,树高精度分别提高1.7%,6.4%,1.8%和0.3%。验证了改进方法的有效性,同时区域内植被混杂程度越高改进算法的提取效果越好。Airborne laser data(Light Detection and Ranging,LiDAR)presents challenges in distinguishing between ground and grassland,and visible light vegetation indices are inadequate for differentiating between shrub and tree layers.Therefore,this study proposes the construction of a multi-band information image that integrates LiDAR point cloud data and RGB vegetation indices.The approach integrates multi-band information from LiDAR point cloud data and vegetation indices to create an enhanced image.The fine-grained canopy height model(CHM)is generated using laser point cloud data.Simultaneously,a high-resolution digital orthophoto image is created using unmanned aerial vehicle imagery data.Among the evaluated vegetation indices,the Differential Enhanced Vegetation Index(DEVI)was the most suitable and was fused with the CHM.Subsequently,the CHM+DEVI fused images were reconstructed to eliminate erroneous values.Training samples were constructed,and the classification regression tree algorithm was employed to segment the ground range and adaptively extract vegetation,such as trees,shrubs,and grasslands.Within the tree areas,the local maximum algorithm was applied to detect tree vertices,which served as foreground markers;meanwhile,the non-tree regions were assigned as background markers.The segmentation results were obtained using watershed transformation,and the accuracy of the extracted vegetation information was analyzed by comparing it with ground-truth data.The evaluation results demonstrate the superior performance of the proposed improved algorithm,with the overall recall rate,precision rate,and accuracy F1 score increasing by 3.2%,3.9%,and 3.5%,respectively.Moreover,the accuracy of tree height measurements exhibited improvements of 1.7%,6.4%,1.8%,and 0.3%in the four quadrats.The effectiveness of the improved method was verified,and the higher the degree of vegetation mixing in the region,the better the extraction effect of the improved algorithm.

关 键 词：激光雷达 无人机影像 差异增强植被指数 形态学重建 标记分水岭算法 

分 类 号：TP394.1[自动化与计算机技术—计算机应用技术] TH691.9[自动化与计算机技术—计算机科学与技术]