板栗树红蜘蛛虫害无人机高光谱遥感监测研究

Hyperspectral Remote Sensing Monitoring of Chinese Chestnut Red Mite Insect Pests in UAV

为了快速、高效、无损监测板栗树的红蜘蛛病虫害,以实地采集的板栗树局部感染明显叶片、感染轻重不均匀叶片、恢复中的感染叶片及不同感染程度叶片为研究对象,利用UHD185型高光谱相机和数码相机获取各种叶片的高光谱图像和RGB图像,以RGB图像为参考,选择各种叶片的感兴趣区,在高光谱图像上提取感兴趣区的光谱曲线,通过微分运算提取光谱曲线的绿峰、红谷、低位、红边、高位、高肩6种光谱特征及特征波长,利用大量实测数据分析板栗树叶各个光谱特征及特征波长随红蜘蛛病虫害危害程度的叶片级变化规律,得到识别红蜘蛛病虫害最佳的光谱特征。利用无人机(Unmanned aerial vehicle,UAV)搭载UHD185型相机,获取了实验区高光谱影像。结果表明,监测板栗树红蜘蛛病虫害危害程度的最佳光谱特征为红边和低位,其与红蜘蛛病虫害的决定系数均超过0.6,当发生轻度红蜘蛛病虫害时,红边波长和低位波长出现"蓝移",说明无人机高光谱遥感系统具有早期发现红蜘蛛病虫害的能力,可为板栗树红蜘蛛病虫害的及时治理提供科学依据。

In order to quickly,nondestructively and efficiently monitor the diseases and insect pests of red mite in chestnut trees,the hyperspectral images and RGB images of each leaf were collected from locally infected leaves,unevenly infected leaves,recovered infected leaves and infected leaves with different degrees of infection by using UHD185 hyperspectral camera and digital camera. The RGB images were used as reference to select the regions of interest( ROI) of each leaf. The spectral curves of ROI were extracted from the hyperspectral images, and six spectral features and characteristic wavelengths of green peak,red valley,low position,red edge,high position and high shoulder of the spectral curves were extracted by differential operation. A large number of measured data were used to analyze the leaf-level variation of the spectral characteristics and characteristic wavelengths of chestnut leaves with the damage degree of red mite pests and diseases,so as to obtain the best spectral characteristics for identifyingred mite pests. After that,the hyperspectral image of the experimental area was obtained by using the UHD185 camera carried by the unmanned aerial vehicle( UAV). The results showed that the best spectral characteristics of monitoring the harm degree of Chinese chestnut red mite were low position and red edge,and the coefficient of determination between red mite and disease and pest exceeded 0. 6. The blue shift of characteristic wavelength could be found in mild red mite pest by using these two characteristics,which proved that the red mite pest could be found in 14 ~ 21 d before the large-scale occurrence of red mite in chestnut tree by UAV hyperspectral remote sensing. The research result can provide a scientific basis for the timely management of diseases and insect pests.