| 摘要: |
| 【目的】利用单次无人机飞行生成的正射影像和数字高程模型(DSM)对果树进行精准识别。【方法】首先利用无人机正射影像计算五种归一化植被指数,并讨论五种植被指数提取植被区域的精度,选用结果最好的植被指数对研究区植被进行提取；之后根据影像的空间分辨率和已知的果树直径范围对果树进行初识别确定果树实际位置和半径；再将识别到的果树叠加到DSM中,利用果树在DSM中最大值和果树临近区域DSM最小值求取果树高度；最后根据果树高度范围对初识别的果树进行再识别,提高果树识别精度。【结果】将该方法运用在美国加利福利亚州弗雷斯诺县里德利市郊区的一个果园进行研究,发现MRENDVI植被指数对研究区内植被提取精度最高；同时利用提取植被区域后影像和果树冠层的直径范围对果树进行初识别的精度为94.8%；利用果树初识别影像与DSM影像结合求取果树高度后,根据对果树高度范围对果树进行再识别后,果树的识别精度提高了5%,达到了99.8%。【结论】该方法原理简单,对果园果树识别有较高的精度,有效消除了基于果树冠层直径范围识别果树过程中,果园周围其他树木和果园内部草丛被误识别为果树的情景,有较高的普适性。 |
| 关键词: 果树识别 无人机影像 数字高程模型 果树高度 |
| DOI： |
| 分类号: |
| 基金项目:中国农业科学院基本业务研究费专项——农业智能机器人技术与装备研发(Y2018YJ14) |
|
| Accurate Detection of Fruit TreesUsing A Set of Unmanned Aerial Vehicle (UAV) Imageries |
|
Lifugen
|
|
The Institute of Agricultural Resources and Regional Planning (IARRP) of Chinese Academy of Agricultural Sciences (CAAS)
|
| Abstract: |
| [Purpose] Fruit trees detection is essential for monitoring the growth of fruit trees, estimating the yield of orchards and planting management for orchards. We proposed a method to accurately detect fruit trees using a set of Unmanned Aerial Vehicle (UAV) Imageries (including a Orthophoto Imagery and digital elevation model (DSM)). [Method] Firstly, five normalized vegetation indices are calculated using a UAV orthophoto image. We discussed the accuracy of extracting vegetation area from five vegetation indices and choosing the best vegetation area that extracted by one of the five vegetation indexes in the study area. Sencondly, the actual position and radius of fruit trees are determined by preliminary detection of fruit trees based on the spatial resolution of images and the specified canopy diameter range of fruit trees. Thirdly, combing preliminary detection of fruit trees and DSM, the height of fruit trees is calculated by the difference of the maximum value of fruit trees area and the minimum value of adjacent area of fruit trees in DSM. Finally, according to the height range of the fruit tree and the preliminary detection of fruit tree, we accurately detect the fruit trees. [Result] The method was applied to an orchard in the suburb of Ridley City, Fresno County, California, USA. It was found that Red Edge Normalized Vegetation Index (MRENDVI) had the highest accuracy in extracting vegetation area of this study. Then we used the image of vegetation area and the specified diameter range of fruit tree canopy to preliminarily detect the fruit trees. The accuracy of preliminary detection of fruit trees is 94.8%. Finally we combine the preliminary detection of fruit trees and DSM image to calculate the fruit trees height, and use the height range of fruit tree to accurately detect the fruit trees. The accurcy of accurate detection of fruit trees increased by 5% to 99.8%. [Conclusion] The method, which is simple in principle, has high accuracy in the fruit trees detection. It effectively eliminates the error detection due to the other trees around the orchard and grass in the orchard according to specified diameter range of fruit tree canopy. This proves that the method has high universality. |
| Key words: fruit tress detection UAV imagery digital elevation model (DSM) fruit trees height |