摘要
基于倾斜影像提取单木参数是当前无人机在林业研究中的热点,以银杏(Ginkgo biloba L.)阔叶林分为研究对象,利用无人机倾斜摄影数据,采用局部最大值法进行了3种探测窗口的单木顶点识别和单木树高提取,并分别进行了精度验证;比较了种子区域增长算法和标记控制分水岭算法提取冠幅的精确度。结果表明:1)在3m×3m,5m×5m,7m×7m探测窗口下,银杏树顶点识别的F得分分别为0.87,0.88,0.83,5m×5m窗口识别效果最好,其树高预测拟合方程中R2达到了0.99,RMSE为1.91m;2)在预测冠幅相对误差为30%的情况下,种子区域增长算法和标记控制分水岭算法提取树冠面积的正确率分别为73.14%和63.43%,对于建立预测与实测冠幅的线性回归关系中,两种算法的R2分别为0.98和0.97,RMSE分别为1.79m^(2)和2.44m^(2),总体上种子区域增长算法的树冠分割精度较标记控制分水岭算法高。研究指出了无人机倾斜摄影技术在对银杏阔叶林进行自动化、精准化单木识别与分割上具有可行性,在森林资源调查中具有一定的应用潜力。
Extracting single tree parameters based on UAV tilting photography data is a hot topic in forestry research.Taking ginkgo(Ginkgo biloba L.)broad-leaved forest as the research object,based on UAV tilting photography data,this paper used the local maximum based algorithm to identify the single tree vertex and extract the single tree height under three detection windows(3m×3m、5m×5m、7m×7m),and the accuracies of recognition and extraction were verified respectively.Then,this paper applied the seed region growth algorithm and marker-controlled watershed algorithm for tree canopy extraction,and the extraction accuracies of two algorithms were compared.The results showed that:1)Under the three single tree detection windows,the F scores of ginkgo tree vertex recognition were 0.87,0.88 and 0.83 respectively.The 5×5m window had the best recognition effect,and in its tree height prediction fitting equation,R2 reached 0.99 and RMSE was 1.91m;2)When the relative error threshold of predicted canopy was 30%,the accuracies of seed region growth algorithm and marker-controlled watershed algorithm in extracting crown area were 73.14%and 63.43%respectively.In establishing the linear regression relationship between predicted and measured canopy,the R2 of the two algorithms were 0.98 and 0.97,and the RMSE were 1.79m^(2) and 2.44m^(2) respectively.In general,the single tree canopy segmentation accuracy of seed region growth algorithm was higher than that of marker-controlled watershed algorithm.This study points out that the UAV tilting photography technology is feasible in the automatic and accurate single tree identification and segmentation of ginkgo broad-leaved forest,thus it has great application potential in forestry investigation.
作者
陈周娟
程光
卜元坤
黄维
陈佳卉
李卫忠
CHEN Zhoujuan;CHENG Guang;BU Yuankun;HUANG Wei;CHEN Jiahui;LI Weizhong(College of Forestry,Northwest A&F University,Yangling,Shaanxi 712100,China;Shaanxi Academy of Forestry,Xi′an710082,China)
出处
《林业资源管理》
北大核心
2022年第1期132-141,共10页
Forest Resources Management
基金
陕西省林业科技创新计划专项“陕西省典型低质低效林精准经营关键技术研究与示范(SXLK2021-0208)”
陕西省林业科学研究重大项目(SHLY-2018-02)。
关键词
倾斜摄影
树高
树冠
局部最大值
区域增长算法
分水岭算法
tilting photography
tree height
canopy
local maximum algorithm
region growth algorithm
watershed algorithm
