刚竹毒蛾危害下的毛竹叶片光谱特征及虫害等级检测研究

Spectral Characteristics of Moso Bamboo Leaves Damaged by Pantana Phyllostachysae Chao and Monitoring of Pest Rating

探讨刚竹毒蛾危害下的毛竹叶片光谱特征可为建立竹林生态安全监测体系提供重要的理论指导。相比于传统的多光谱数据,高光谱遥感能够准确探测不同刚竹毒蛾危害等级间寄主光谱的细微变化。然而,当前有关此方面的研究甚少,其寄主的光谱变化机理还有待进一步总结。为此,基于实测的552条竹叶光谱,分析了健康、受刚竹毒蛾危害、小年叶片之间的光谱差异,选择可反映其健康状况的特征变量,并利用XGBoost模型建立了叶片尺度的刚竹毒蛾危害检测模型。研究结果显示:(1)随着虫害等级的上升,受害叶片在可见光范围内的反射率逐渐出现“绿低红高”的特征,其近红外波段的反射率则不断降低,而短波红外的反射率则明显高于健康叶片,尤其在两个水汽吸收波段(1450和1940 nm)的差异最为明显;(2)小年叶片于可见光-近红外波段的反射率显著高于健康、受害叶片;(3)根据不同受害类型叶片的光谱特征可知,较之健康叶片,缺刻型叶片的光谱并未出现太大的变化,红褐色病斑型叶片在红光波段的反射率出现了一定程度的上升,灰白色病斑型叶片则已经完全失去了植被的基本光谱特征;(4)根据XGBoost模型给出的变量重要性排序可知,各特征变量的贡献度依次为PRI(光化学反射率指数)>FDVI 576,717(植被健康程度评估指数)>NPCI(归一化色素叶绿素指数)>DSWI(疾病水胁迫指数)>VOG 1(红边指数1)>RVSI(红边植被胁迫指数)>NDWI(归一化差值水分指数);(5)模型对刚竹毒蛾危害识别的总平均精度为74.39%,其中健康叶片的识别精度达到了94.55%,轻度危害叶片为74.93%,重度危害为84.12%,小年叶片则为71.10%,而中度危害叶片的识别精度较差,仅为33.48%。

Understanding the spectral characteristics of moso bamboo leaves damaged by Pantana phyllostachysae Chao can provide theoretical guidance for developing applicable and effective technologies to monitor the ecological safety of the bamboo forest.Compared with the traditional multispectral data,hyperspectral remote sensing can sense the subtle changes of host spectrum among different severity of Pantana phyllostachysae Chao.However,the related researches were still rare,and the spectral change mechanism of the host needs to be further summarized.Therefore,this study analyzed the spectral differences among healthy,damaged and off-yearmoso bamboo leaves based on 552 field measured spectrums.The characteristic variables that can act as indicators of leaves health status were selected.Finally,the model for monitoring the damage of leaves caused by Pantana phyllostachysae Chao was established using the XGBoost algorithm.The results show that:(1)with the increase of pest damage,the reflectance of damaged leaves gradually appeared“green low and red high”in visible-band,while the reflectance noticeably decreased in near-infrared band,and the reflectance of damaged leaves in shortwave infrared band was significantly higher than that of healthy leaves,especially in the two typical water vapor absorption bands(1450 and 1940 nm);(2)the reflectance of off-year leaves in visible and near infrared bands was significantly higher than healthy and damaged leaves;(3)the spectral characteristics of indentation-only leaves only slightly changed in comparison with healthy leaves,while the red band reflectance of leaves with red-brown disease spots increased to some extent,andthe leaves with gray-white disease spots completely lost the basic spectral characteristics of vegetation;(4)according to the feature importance score determined by the XGBoost algorithm,the contribution of each characteristic variable was PRI>FDVI 576,717>NPCI>DSWI>VOG 1>RVSI>NDWI;(5)the overall average accuracy of the model to detect the damage by the pest was 74.39%,and the accuracy for healthy,mild damaged,severe damaged,off-year,and moderate damaged leaves was 94.55%,74.93%,84.12%,71.10%,and 33.48%,respectively.