Fast R-CNN深度学习和无人机遥感相结合在松材线虫病监测中的初步应用研究

松材线虫病因其破坏性强、传播速度快和防治难度大等特点,严重威胁着我国的松林资源.及时发现、定位和清理病死松树是控制松材线虫病蔓延的有效手段.本研究利用小型无人机获得松材线虫病疫点的可见光和多光谱的航摄影像.根据松树针叶颜色变化,将松材线虫Bursaphelenchus xylophilus侵染的松树分为病树和枯死树两种类型.将无人机遥感正摄影像图切割成瓦片图,根据不同植被指数的特征差异,筛选出含病树和枯死树的瓦片图.训练Fast R-CNN深度学习框架形成最终模型,通过模型运算获得病枯死松树的分布地图及坐标点位置.研究结果显示Fast R-CNN深度学习和无人机遥感相结合能有效识别出病树和枯死树,正确率分别达到90％和82％,漏检率分别为23％和34％,可为大面积监测松材线虫病的发生现状和流行动态、评估防控效果和灾害损失提供技术支撑.

The preliminary application of the combination of Fast R-CNN deep learning and UAV remote sensing in the monitoring of pine wilt disease

Pine wilt disease (PWD) poses a huge threat to pine wood forest because of its destructiveness, rapid spread and difficult prevention and treatment. Finding, locating and cleaning the disease or death pine trees in time was an effective means to control the spread of PWD. This paper used a small fixed-wing unmanned aerial vehicle (UAV) acquisition platform to obtain visible light orthophotograph image and multispectral orthophotograph image of disease areas of PWD. According to the color change of the pine needles, the pine trees infected by Bursaphelenchus xylophilus were divided into two categories of diseased trees and dead trees. The UAV remote sensing orthophoto images were cut into tile maps, and the tile maps containing diseased trees or dead trees were screened according to the difference in characteristics of different vegetation index.Training Fast R-CNN depth learning frame formed a final model, and the distribution maps and location coordinates of diseased trees and dead trees were obtained through the final model operation. The results showed that the combination of Fast R-CNN deep learning and UAV remote sensing could effectively identify the diseased tree and dead tree, the accuracy rate were 90% and 82%, the leak detection rate was 23% and 34%. This method could provide technical support for the research and control of PWD, including the large-scale monitoring of the occurrence status and epidemic dynamics, evaluation control effects and disaster losses.