| 引用本文： |
-
林鑫,潘婷,刘震峰,黄钰清.无人机地理空间视频自动识别松材线虫病疫区枯立木研究[J].广西科学,2026,33(3):556-568. [点击复制]
- LIN Xin,PAN Ting,LIU Zhenfeng,HUANG Yuqing.Automatic Recognition of Standing Dead Trees in Pine Wood Nematode Epidemic Areas from UAV Geospatial Videos[J].Guangxi Sciences,2026,33(3):556-568. [点击复制]
|
|
| 摘要: |
| 针对松材线虫病疫区对枯立木精准、高效的监测需求，为解决传统正射影像技术采集条件严苛、处理流程冗长、成本偏高等问题，本研究以正射影像人工判读结果为对照，选取DJI Mavic 3E(M3E)、DJI Mavic 3M(M3M)、DJI Mavic 4E(M4E)3款常用行业级无人机(UAVs)为作业平台，在广西壮族自治区柳州市三江侗族自治县、贺州市钟山县和岑溪市松材线虫病疫区开展手动飞行与自动航线飞行两种模式下的地理空间视频识别枯立木试验，并引入YOLOv8-LSK算法实现对枯立木的自动识别与定位，探讨航高、云台角度、地形起伏、无人机机型及影像数据源类型对识别准确率与定位偏差的影响。结果显示，视频识别的数据采集与处理总耗时大幅少于正射影像，三江侗族自治县M4E手动飞行比正射影像减少89.57%，钟山县自动航线飞行300 m航高比正射影像减少61.63%。M4E凭借21 m/s的飞行速度实现最优采集效率，M3M同步采集多光谱数据，耗时仍远低于同机型正射影像。不同地形下最优参数组合(航高50—150 m、云台角度90°、可见光数据源、M4E机型)的枯立木识别准确率为88.9%—95.8%。视频识别的定位偏差在低航高垂直视角下的平均值为0.69—1.98 m，符合森林资源调查常规精度要求(±3.00 m)；手动飞行模式定位偏差平均值为3.28 m，略超常规精度要求阈值，但其识别准确率在高航高下优于自动航线飞行，人为干预可以弥补设备性能不足，平衡调查效率与识别精度。地理空间视频识别松材线虫病枯立木效率优势突出，识别精度可控，定位精度达标，可以应用于松材线虫病疫区枯立木的监测和调查。 |
| 关键词: 无人机 地理空间视频 目标识别 枯立木 松材线虫病 |
| DOI：10.13656/j.cnki.gxkx.20260708.010 |
| 投稿时间：2026-04-28修订日期：2026-05-22 |
| 基金项目:广西自筹经费林业科技项目“基于无人机遥感的松材线虫病受害木自动识别研究”(2023GXZCLK71)资助。 |
|
| Automatic Recognition of Standing Dead Trees in Pine Wood Nematode Epidemic Areas from UAV Geospatial Videos |
|
LIN Xin, PAN Ting, LIU Zhenfeng, HUANG Yuqing
|
| (Guangxi Forest Inventory & Planning Institute, Nanning, Guangxi, 530011, China) |
| Abstract: |
| To address the need for precise and efficient monitoring in pine wood nematode epidemic areas and overcome the limitations-strict acquisition conditions,lengthy processing workflows,and high costs-of conventional orthophoto techniques,this study used manual interpretation of orthophotos as the control.Three commonly used industrial grade Unmanned Aerial Vehicles (UAVs) DJI Mavic 3E (M3E),DJI Mavic 3M (M3M),and DJI Mavic 4E (M4E) were deployed as operational platforms.Geospatial video-based recognition experiments of standing dead trees were conducted under both manual and automatic flight route modes in the epidemic areas of Sanjiang Dong Autonomous County,Zhongshan County,and Cenxi City in Guangxi.The YOLOv8-LSK algorithm was employed for automatic recognition and positioning,and the influences of flight altitude,pitch angle,terrain relief,UAV model,and image data source type on recognition accuracy and positioning deviation were investigated.Results showed that the total time for data acquisition and processing of video recognition was substantially shorter than that of orthophoto.In Sanjiang Dong Autonomous County,manual flight with M4E took only 89.57% of the orthophoto time,and in Zhongshan County,automatic flight route at 300 m altitude took only 61.63% of the orthophoto time.M4E achieved the best acquisition efficiency owing to its 21 m/s flight speed.Even when M3M synchronously collected multispectral data,its time consumption remained far lower than the orthophoto of the same model.Across different terrains,the optimal parameter combination (flight altitude of 50—150 m,pitch angle of 90°,visible-light data source,and M4E) yielded the standing dead tree recognition accuracy of 88.9%—95.8%.The average positioning deviation of video recognition at low flight altitude and vertical viewing angle ranged from 0.69 m to 1.98 m,meeting the conventional accuracy requirement of forest resource surveys (±3.00 m).The average positioning deviation of the manual flight mode was 3.28 m,slightly exceeding the conventional accuracy threshold.However,the recognition accuracy of the manual flight mode was superior to that of the automatic flight route at high flight altitudes.Human intervention can compensate for limited equipment performance and help balance survey efficiency with recognition accuracy.Geospatial video-based recognition of standing dead trees killed by pine wood nematodes demonstrates outstanding efficiency,controllable recognition accuracy,and acceptable positioning accuracy,and can thus be applied to the monitoring and surveying of standing dead trees in pine wood nematode epidemic areas. |
| Key words: UAV geospatial video target recognition standing dead tree pine wood nematode |
