应用无人机影像提取毛竹林立竹度  

Extracting Phyllostachys edulis Stand Density with UAV Images

作　　者：杨樟平 曹碧凤 谢巧雅 邓洋波[2] 刘健[2] 余坤勇[1] Yang Zhangping;Cao Bifeng;Xie Qiaoya;Deng Yangbo;Liu Jian;Yu Kunyong(Fujian Agriculture and Forestry University,Fuzhou 350002,P.R.China;Yong’an Forestry Bureau)

机构地区：[1]福建农林大学林学院 [2]福建农林大学,福州350002 [3]永安市林业局

出　　处：《东北林业大学学报》2021年第7期66-71,共6页Journal of Northeast Forestry University

基　　金：福建高校产学项目(2019N5012)、(2020N5003);福建农林大学科技计划创新项目(KFA18130A)。

摘　　要：采用无人机航拍影像,依据毛竹冠幅几何特点,依据面向对象多尺度分割原理构建毛竹株数量识别单元,比对分析K邻近法(KNN)、支持向量机(SVM)、随机森林(RF)的提取效果,确定毛竹立竹度提取的最优算法。结果表明:1株毛竹几何形状趋近圆形,2株趋近于长椭圆形,3株以上近似于长条形,与图像中毛竹的冠层形状特征与毛竹识别单元类型具有一致性。验证结果KNN、SVM、RF提取立竹度总体精度平均值分别为90.70%、89.64%、94.56%,对应的Kappa系数分别为0.859 7、0.830 6、0.945 6,其中RF的精度最高。整体上,基于毛竹立竹度识别单元的构建,结合RF分类方法比其他两类分类方法更具优势,实现了有效的毛竹林立竹度提取。UAV aerial imagery is used,according to the geometric expression characteristics of Phyllostachys edulis crown,based on the object-oriented multi-scale segmentation principle,the number of P.edulis identification units is constructed,and K-Nearest Neighbor(KNN),Support Vector Machine(SVM)and random forest(RF)extraction effects to determine the optimal algorithm for extracting bamboo shoots.The geometric shapes of 1 bamboo were close to round,2 were close to oblong,and more than 3 were close to long.It is consistent with the canopy shape features of P.edulis in the image and the types of recognition units of P.edulis.The overall average accuracy of KNN,SVM,and RF extraction of bamboo shoots are 90.70%,89.64%and 94.56%,corresponding to the Kappa coefficients were 0.8597,0.8306 and 0.9456,of which RF accuracy is the highest,respectively.In summary,from the construction of the recognition unit of bamboo stand bamboo motility,combined with the RF classification method,it has more advantages than the other two types of classification methods,and high precision bamboo stand extraction.

关 键 词：毛竹 立竹度 面向对象多尺度分割 识别单元 最优分类方法 

分 类 号：S757.2[农业科学—森林经理学]