基于高光谱和机载LiDAR技术的云南松受云南切梢小蠹危害程度分类诊断研究

以石林县黑龙潭片区常受云南切梢小蠹危害的云南松为研究对象，人工调查120株云南松冠层枯稍率，并记录其坐标位置。利用无人机搭载高光谱成像仪采集研究区云南松高光谱数据，根据样本位置提取其冠层光谱反射率，分析冠层光谱，并利用小波变换法提取了16个光谱特征。基于样本冠层枯稍率与光谱特征数据，采用BP神经网络法训练分类诊断模型，以此作为研究区云南切梢小蠹对云南松危害程度分类诊断模型。利用无人机搭载LiDAR系统采集研究区云南松LiDAR数据，根据归一化切割方法，对云南松单株树冠进行分割提取，将分割提取结果与高光谱数据融合提取单株云南松冠层光谱。最终利用训练好的诊断模型分类诊断研究区云南松受云南切梢小蠹危害程度并将结果可视化。结果表明：共分割提取出11029株云南松，分类结果为健康10142株、轻度危害490株、中度危害266株、重度危害131株。选用120株样本对分类结果进行精度验证，总体分类精度为90.83%，分类精度较高。

Classification Diagnosis on the Damage Degree of Tomicus yunnanensis to Pinus yunnanensis Based on Hyperspectral and Airborne LiDAR

Taking Pinus yunnanensis in the Heilongtan area of Shilin County, which is often harmed by Tomicus yunnanensis, as the research object, the canopy dieback rate of 120 P. yunnanensis was manually investigated and their locations were recorded. A UAV equipped with a hyperspectral imager was used to collect the hyperspectral data of P. yunnanensis in the study area, and the canopy spectral reflectance was extracted according to the sample location, the canopy spectrum was analyzed, and 16 spectral features were extracted using the wavelet transform method. Based on the sample canopy withering rate and spectral feature data, the BP neural network method was used to train the classification diagnosis model, which was used as the classification diagnosis model for the damage degree of T. yunnanensis to P. yunnanensis in the study area. UAV-mounted LiDAR system was used to collect LiDAR data of P. yunnanensis in the study area. According to the normalized cutting method, the individual canopy of P. yunnanensis was segmented and extracted, and the segmentation and extraction results were fused with hyperspectral data to extract the canopy spectrum of individual P. yunnanensis. Finally, the trained diagnostic model was used to classify and diagnose the damage degree of P. yunnanensis in the study area by T. yunnanensis and visualize the results. The results showed that a total of 11029 P. yunnanensis strains were divided and extracted. The classification results were 10142 healthy strains, 490 mildly damaged strains, 266 moderately damaged strains, and 131 severely damaged strains. 120 samples were selected to verify the accuracy of the classification results.The overall classification accuracy was 90.83%, and the classification accuracy was high.