无人机多光谱的竹林病虫害监测——以福州市晋安区沙溪村典型受灾区为例

Research on Pests and Diseases Monitoring of the Bamboo Forest Based on UAV Multi-spectrum-A Case Study of Shaxi Village,a Typical Disaster-Stricken Area in Jin an District,Fuzhou City

以竹林病虫害典型受灾区——福建省福州市晋安区寿山乡沙溪村为研究对象,结合研究区气象历史数据、灾害记录,实地调查数据等,开展高分辨率无人机多光谱竹林病虫害监测.利用互信息(mutual information,MI)衡量特征变量(NDGI、VARI、Blue、Green、Red)与竹林受病虫害影响情况之间的相互依赖程度,筛选适合识别竹林病虫害的光谱特征,选取最优的监测数据方案;然后基于随机森林算法(Random Forest,RF)对研究区竹林病虫害监测进行精度评估.结果表明:(1)沙溪村竹林典型受灾区中,受灾竹子占竹林总面积的46.66%,其中重度受灾竹子占竹林总面积的33.4%,且分布较广;(2)抗大气植被指数(vegetation atmospherically resistant index,VARI)对毛竹受病虫害的影响程度十分敏感,MI值达3.110,是适用于监测竹林病虫害的光谱特征;(3)相对于仅用蓝、绿和红波段进行监测,对采用蓝、绿、红波段Bands&VARI的数据方案下,各受灾程度的识别精度整体较高;基于随机森林算法对竹林病虫害进行监测的总体分类精度可达83.58%,Kappa系数为0.72.

In this paper,Shaxi Village,Shoushan Township,Jin’an District,Fuzhou City,Fujian Province,a typical disaster-stricken area affected by pests and diseases of the bamboo forest,was taken as the research object.On one hand,in combination with the historical meteorological data,disaster records and field survey data of the study area,pests and diseases monitoring of the bamboo forest based on high-resolution UAV multi-spectrum was carried out,the degree of interdependence between characteristic variables(NDGI,VARI,Blue,Green,Red)and the impact of pests and diseases on the bamboo forest were measured by using mutual information(MI),the spectral characteristics suitable for identifying pests and diseases in the bamboo forest was screened,and the optimal monitoring data scheme was selected.On the other hand,the monitoring accuracy of the bamboo forest pests and diseases in the study area was evaluated based on random forest(RF)algorithm.The results are shown as follows.Firstly,in the typical disaster-stricken area of the bamboo forest in Shaxi Village,the affected bamboo accounted for 46.66%of the total area of the bamboo forest,of which the severely affected bamboo accounted for 33.4%,and the distribution was wide.Secondly,vegetation atmospherically resistant index(VARI)is highly sensitive to the impact of bamboo pests and diseases,with a MI value of 3.110,which is a spectral feature suitable for monitoring pests and diseases of the bamboo forest.Finally,compared with monitoring only with blue,green and red bands,the identification accuracy of each degree of disaster is higher under the data scheme using blue,green and red bands&VARI.Generally speaking,the overall classification accuracy of the bamboo forest pests and diseases monitoring based on random forest(RF)algorithm can reach 83.58%,and the Kappa coefficient is up to 0.72.