| 基于无人机影像自动检测冠层果的油茶快速估产方法 |
| |
| 引用本文： | 严恩萍,棘玉,尹显明,莫登奎.基于无人机影像自动检测冠层果的油茶快速估产方法[J].农业工程学报,2021,37(16):39-46. |
| |
| 作者姓名： | 严恩萍 棘玉 尹显明 莫登奎 |
| |
| 作者单位： | 1. 中南林业科技大学林业遥感大数据与生态安全湖南省重点实验室，长沙 410004；2. 中南林业科技大学南方森林资源经营与监测国家林业与草原局重点实验室，长沙 410004；3. 中南林业科技大学林学院，长沙 410004 |
| |
| 摘 要： | 快速准确的产量估算对油茶经营管理和可持续发展具有重要意义。该研究针对油茶快速估产的应用现状,提出一种基于无人机影像自动检测冠层果的方法用于油茶快速估产。首先借助无人机航拍影像,通过随机抽样选取120株油茶树进行无人机近景摄影和人工采摘称量;然后利用Mask RCNN(Mask Region Convolutional Neural Networks)网络开展基于近景影像的油茶冠层果自动检测与计数;采用线性回归和K最邻近建立冠层果数与单株果数之间的关系,同时结合研究区典型样木株数和平均单果质量,构建基于冠层果自动检测的估产模型。结果表明:1)无人机超低空近景影像结合Mask RCNN网络能够有效检测不同光照条件油茶果,平均F1值达89.91%;2)同传统卫星遥感相比,基于无人机近景摄影的冠层果自动检测在作物产量估测方面显示出明显优势,Mask RCNN网络预测的冠层果数与油茶样木单株果数之间具有良好的一致性,拟合决定系数R2达0.871;3)结合线性回归和K最邻近构建的模型估产精度均较高,拟合决定系数R~2和标准均方根误差NRMSE(Normalized Root Mean Square Error)分别在0.892～0.913和28.01%～31.00%之间,表明基于无人机影像自动检测冠层果的油茶快速估产是一种切实可行的方法。研究结果可为油茶快速估产和智能监测提供参考。
|
| 关 键 词： | 无人机 自动检测 油茶 快速估产 冠层果 |
| 收稿时间： | 2021/7/14 0:00:00 |
| 修稿时间： | 2021/7/14 0:00:00 |
Rapid estimation of camellia oleifera yield based on automatic detection of canopy fruits using UAV images |
| |
| Yan Enping,Ji Yu,Yin Xianming,Mo Dengkui.Rapid estimation of camellia oleifera yield based on automatic detection of canopy fruits using UAV images[J].Transactions of the Chinese Society of Agricultural Engineering,2021,37(16):39-46. |
| |
| Authors: | Yan Enping Ji Yu Yin Xianming Mo Dengkui |
| |
| Institution: | 1. Key Laboratory of Forestry Remote Sensing Big Data & Ecological Security for Hunan Province, Central South University of Forestry & Technology, Changsha 410004, China; 2. Key Laboratory of State Forestry and Grassland Administration on Forest Resources Management and Monitoring in Southern Area, Central South University of Forestry & Technology, Changsha 410004, China; 3. College of Forestry, Central South University of Forestry & Technology, Changsha 410004, China |
| |
| Abstract: | Abstract: Rapid and accurate yield estimation is of great significance to the management and sustainable development of Camellia oleifera production. The quantity and single fruit weight of Camellia fruits are crucial indicators representing the Camellia yield. Therefore, a highly efficient and accurate monitoring of the quantity and single fruit weight of Camellia fruits can contribute to saving labor, material, and financial resources, as well as timely decision-making. Unmanned aerial vehicle (UAV) remote sensing has presented a high spatial resolution, fast data acquisition, and simple operation in recent years. An optimal operation period can be selected to obtain high-resolution aerial images, and thereby to realize the crop yield estimation in a large scale using the fruit numbers. However, only a few kinds of research are focused on the estimation of Camellia yield using UAV images and fruit number identification. In this study, a rapid yield estimation of Camellia oleifera was realized via the automatic detection of canopy fruit using UAV images. Firstly, a DJI Mavic 2 PRO UAV platform and Hasselblad L1D-20C camera were utilized to obtain UAV aerial images in the study area. 120 Camellia oleifera trees were selected by random sampling for close-up UAV shooting, and manual picking, and weighing. A Mask RCNN framework was then employed to automatically detect and count the canopy fruits in the sample trees using UAV close-up images. Finally, two common Linear Regression (LR) and K-Nearest Neighbor (KNN) were used to build the relationship between the predicted fruit numbers of tree canopy and the measured. A yield estimation model was thus constructed using automatic detection of canopy fruits, according to the total numbers of sample trees and the average weight of single fruit. The results showed that: (1) There was an excellent performance of crop yield estimation via the automatic detection of canopy fruits using close-range photography of UAV, compared with the traditional. (2) UAV ultra-low-altitude close-up images combined with Mask RCNN network effectively detected the camellia oleifera fruits under different lighting conditions, with an average F1 value of 89.91%. (3) There was well consistency between the predicted fruit numbers of tree canopy identified by Mask RCNN network and the measured, with R2 higher than 0.871. (4) The yield estimation results showed that the combined LR/KNN models presented a higher accuracy of yield estimation with R2 and NRMSE ranging from 0.892 to 0.913, and 28.01% to 31.00%, respectively. (5) The total estimated yields of sample trees were 66.950 and 64.600 kg using LR and KNN models, respectively, where the estimation accuracy was 96.43% and 93.04%, respectively. Consequently, the rapid yield estimation of Camellia oleifera sample trees can be achieved using automatic detection of canopy fruits from UAV images. The finding can provide highly versatile and great potential for rapid yield estimation and intelligent monitoring of the crops or trees in large areas. |
| |
| Keywords: | UAV automatic detection Camellia oleifera rapid yield estimation canopy fruit |
| 本文献已被 CNKI 等数据库收录！ |
| 点击此处可从《农业工程学报》浏览原始摘要信息 |
| 点击此处可从《农业工程学报》下载免费的PDF全文 |
|
