基于无人机影像的山核桃单木检测及冠幅与树高的提取  

Research on single tree detection and crown diameter and tree height extraction of pecan forest based on UAV images

作　　者：郭阳光 夏凯[1] 杨垠辉 冯海林[1] GUO Yangguang;XIA Kai;YANG Yinhui;FENG Hailin(College of Mathematics and Computer Science,Zhejiang A&F University,Zhejiang Provincial Key Laboratory of Forestry Intelligent Monitoring and Information Technology,Key Laboratory of Forestry Perception Technology and Intelligent Equipment,National Forestry and Grassland Administration,Hangzhou 311300,China)

机构地区：[1]浙江农林大学数学与计算机科学学院,浙江省林业智能监测与信息技术研究重点实验室,林业感知技术与智能装备国家林业和草原局重点实验室,杭州311300

出　　处：《林业工程学报》2023年第4期159-166,共8页Journal of Forestry Engineering

基　　金：浙江省自然科学基金联合基金(LQY18C160002);浙江省自然科学基金(LQ20F020005);浙江省“尖兵”计划择优委托项目(2022C02009);国家自然科学基金(32271869)。

摘　　要：针对经济林中树木的生长状况进行调查,有助于农户制定针对性的经营策略,提高经营效率。然而,由于山核桃树通常生长在山地环境下,使用传统的林业调查方法获取树木参数需要的人力资源和时间成本较高,而且在陡峭的山地环境中容易受到地形、植被和气象等因素的干扰。为了解决这一问题,提出了一种新的自动化方法——检测框投影法。该方法基于深度学习的目标检测算法对遥感图像中的树冠进行检测并生成检测框,再依据所得到的检测框获取树木位置和数量,并进一步提取单木的冠幅与树高等参数。在不同环境的山核桃种植林场进行的树冠检测实验结果表明,该方法使用的目标检测算法对山核桃树冠检测的总体平均精度和F1-score分别达到了85.5%与0.84;参数提取方面,在两处不同的山核桃种植林场选取了3处研究样地,并在每处样地选取并实地测量了50棵样本树木的冠幅和树高以验证参数提取精度,结果表明,使用检测框投影法预测冠幅与实测值的均方根误差、平均绝对误差和平均相对误差分别为0.469 m、0.313 m和5.7%,预测树高与实测值的均方根误差、平均绝对误差和平均相对误差分别为0.427 m、0.331 m和6.0%。提出的检测框投影法在山核桃林地环境下可以获得较为准确的树冠检测与参数提取结果,帮助农户制定更加合理和科学的经营策略,提高经营效率,同时也为林业生产的可持续发展提供了参考。As a forestry economic model,economic forests are trees planted to obtain economic benefits.Investigation of the growth status of trees in economic forests can help farmers formulate targeted management strategies and improve management efficiency.However,since the pecan(Carya cathayensis)trees typically grow in mountainous environments,tree parameters are usually obtained using traditional forestry survey methods,which require high human resources and time costs,and are prone to interference from factors such as terrain,vegetation,and meteorology in steep mountain environments.To solve these problems,this study proposed a new automatic method,i.e.,the detection frame projection method.This method is based on a deep learning target detection algorithm to detect the tree crown in remote sensing images and generate a detection frame.Based on the obtained detection frame,the position and number of trees are obtained,and the crown width and tree height parameters of a single tree are further extracted.The experimental results of crown detection in pecan plantation forests in different environments showed that the overall average accuracy and F1 score of the target detection algorithm used in this method for the pecan crown detection reached 85.5%and 0.84,respectively.In terms of parameter extraction,this study selected three research sample plots from two different pecan plantation forests,and selected and measured the crown diameters and tree heights of 50 sample trees in each sample plot to verify the accuracy of parameter extraction.The results showed that,using the detection frame projection method,the root mean square error,average absolute error,and average relative error of the predicted crown diameters and measured values were 0.469 m,0.313 m,and 5.7%,respectively.The root mean square error,average absolute error,and average relative error of the predicted tree height and the measured value were 0.427 m,0.331 m,and 6.0%,respectively.It was indicated that the detection frame projection method proposed in this

关 键 词：山核桃 无人机影像 YOLOv5 树冠检测 参数提取 检测框投影法 

分 类 号：TP701[自动化与计算机技术—检测技术与自动化装置]