关键词：
机载激光雷达;单木分割算法;区域增长;边界控制;能量函数
摘 要：
基于机载激光雷达(Li DAR)技术和单木分割算法提取单株树木信息对于单木结构研究、理解树木生长、森林可持续管理具有重要的意义。本研究以分水岭算法为基础,使用可变窗口的局部最大值算法,并采用分层级的区域增长算法及由高度差和梯度构建的能量函数来分割树冠边界,从而优化树顶提取和相邻树冠的分割结果。以中国南方亚热带森林为研究区,测试了针对不同密度(低、中、高密度)、不同树种(白皮松和桉树)以传统分水岭算法和优化算法对于树顶提取及树冠分割的效果。为检验结果的准确性,在实验中对Li DAR数据和人工测量数据的分割结果进行了对比和验证,结果表明:对树冠顶点探测率而言,优化算法平均探测效果(树冠探测率r=0.90、树冠准确率p=0.84、总体准确率f=0.86)优于传统分水岭算法(r=0.62、p=0.81、f=0.78);对树冠边界探测精度而言,优化算法平均探测效果(R2=0.80、RMSE=0.22 m、RRMSE=12.03%)优于传统分水岭算法(R2=0.68、RMSE=0.28 m、RRMSE=17.45%)。在树冠探测上,桉树的准确率略低于白皮松,但是与白皮松的探测率相差较小,这表明白皮松对于优化的算法具有较好的鲁棒性,受密度等因素影响较小;在冠幅探测上,桉树和白皮松的精确度随着林分密度的增加而增加,且均有良好的表现。本研究是一种在传统分水岭算法基础上优化的算法,可以较好地提高单木信息分割的效果,这对于了解亚热带森林中林木经营管理、树木竞争及资源监测具有重要意义。
译 名：
Individual tree crown separation using the improved watershed method
作 者：
LIU Fangzhou;LIU Hao;YUN Ting;Co-Innovation Center for the Sustainable Forestry in Southern China,Nanjing Forestry University;
关键词：
LiDAR;;individual tree crown segmentation;;region growth;;boundary control;;energy function
摘 要：
Individual tree crown segmentation from Airborne Laser Scanning data is a nodal problem in forest remote sensing.The airborne LiDAR(light detection and ranging) technology and individual tree crown segmentation algorithm have great significance for exploring tree structures,understanding tree growth and benefiting sustainable forest management.Using the watershed algorithm,this study uses a local maximum algorithm with a variable window,a hierarchical region growth algorithm and an energy function constructed from height differences and gradients to segment the tree crown boundary,thereby optimizing the tree top extraction and segmentation results of adjacent crowns.In this study,the subtropical forests of southern China are used as the research area(the Peak Forest Farm in Nanning,China).The effects of traditional watershed algorithms and this proposed algorithm on tree top extraction and crown segmentation under different densities(low,medium,and high density) and different tree species(Pinus bungeana Zucc.and Eucalyptus robusta) are tested.In order to check the accuracy of the results,the segmentation results of the LiDAR data and the manually measured data are compared and verified in the experiments.The results show that,in terms of the average detection rate of treetop detection of the algorithm used in this study(r=0.90,p=0.84,f=0.86) is better than the traditional watershed algorithm(r=0.62,p=0.81,f=0.78).In terms of the average precision of canopy boundary detection,this proposed algorithm(R~2=0.80,R_(MSE)=0.22 m,R_(RMSE)=12.03%) is better than the traditional watershed algorithm(R~2=0.68,R_(MSE)=0.28 m,R_(RMSE)=17.45%).In the treetop detection,the accuracy rate of eucalyptus is slightly lower than that of P.bungeana Zucc.,but the difference between the detection rate of P.bungeana Zucc.and E.robusta is small.In crown width detection,the accuracy of eucalyptus and P.bungeana Zucc.increases with the increase of density,and both tree species have good performance in crown width detection.In the subtropical forest,the proposed algorithm is optimized based on the traditional watershed algorithm,which can improve the estimation accuracy of the individual tree crown segmentation.The result of this study is of great significance for understanding forest cultivation practices,forest competition factor and resource monitoring.
