基于YOLOv3网络无人机影像橘子树识别  

Orange Tree Recognition from UAV Image Based on YOLOv3 Network

作　　者：庞栋栋 刘刚[1,2] 何敬 付饶[1] PANG Dongdong;LIU Gang;HE Jing;FU Rao(College of Earth Sciences,Chengdu University of Technology,Chengdu 610059,China;State Key Laboratory of Geohazard Prevention and Geoenvironment Protection,Chengdu 610059,China)

机构地区：[1]成都理工大学地球科学学院,四川成都610059 [2]地质灾害防治与地质环境保护国家重点实验室,四川成都610059

出　　处：《测绘与空间地理信息》2022年第4期32-36,40,共6页Geomatics & Spatial Information Technology

基　　金：国家自然科学基金项目(41871303,41602355);地质灾害防治与地质环境保护国家重点实验室课题(SKLGP2018Z010);四川省科技计划项目(2021YFG0);四川省自然资源厅(KJ-2021-3)资助。

摘　　要：利用高新遥感技术和先进的目标检测方法快速、准确地提取作物信息对精准农业的发展具有重要意义。为此,提出以高分辨率无人机影像为数据源,用标记数据训练YOLOv3网络,得到最优参数估计,形成针对橘子树识别的神经网络,实现密集的橘子树影像识别。为证明本方法的可靠性,用相同的样本在ENVI5.6平台Deep Learning模块进行实验对比。结果表明,本文方法可以高效、准确地从无人机影像中提取橘子树信息,其识别精度优于ENVI5.6平台Deep Learning模块的识别结果,模型运行稳定、可靠,可以作为统计农作物的可选择性方法。It is important to the development of precision agriculture using advanced remote sensing technology and target detection method to extract crop information rapidly and accurately.Based on this,a high-resolution UAV image was proposed as the data source to train YOLOv3 network with labeled data and to obtain the optimal parameter estimation and form a neural network for orange trees,so as to realize the recognition of the dense orange tree images.In order to prove the reliability of the experiment,the same sample was used for comparison based on the Deep Learning module of ENVI5.6 platform.The results show that the proposed method can efficiently extract orange tree information from UAV images,and its identification accuracy is better than that of Deep Learning module of ENVI5.6 platform.The model is stable and reliable which can be applied as a selective method for crop statistics.

关 键 词：YOLOv3 深度学习 橘子树 无人机影像 精准农业 

分 类 号：P237[天文地球—摄影测量与遥感]