落叶松毛虫危害下落叶松针叶含水率高光谱估算

Hyperspectral estimation of water content in larch needles from Dendrolimus superans

【目的】虫害早期胁迫会使得落叶松针叶含水率等内部生化组分发生微小变化,此时光谱具有明显响应。随着遥感技术的发展,高光谱遥感能够捕捉到这些微小变化,成为监测和识别虫害早期胁迫的重要工具。准确地预测针叶含水率,可为落叶松毛虫虫害早期胁迫预警工作提供实验依据。【方法】以大兴安岭森林虫害爆发区为研究区,获取2021年7月的高光谱数据和针叶含水率数据,通过平滑光谱反射率(SSR)、微分光谱反射率(DSR)和光谱连续小波系数(SCWC)三种光谱特征,结合Findpeaks函数(FP)和连续投影算法(SPA)对光谱降维和提取敏感特征,并利用偏最小二乘回归(PLSR)、随机森林(RF)和支持向量机回归(SVMR)三种算法建立含水率估算模型。【结果】1)SCWC的敏感波段大多集中在2^(2)、2^(3)、2^(4)三个尺度上且比DSR和SSR更加敏感。2)FP-SPA能快速准确地选择敏感光谱特征,减少数据冗余,降低模型复杂度。3)PLSR-SCWC-coif4模型精度最高(R^(2)=0.890 4,RMSE=0.037 5),比DSR和SSR的R^(2)分别提高0.096 3和0.115 3,RMSE分别降低0.007 5和0.017 8;其次是SVMR-SCWC-bior4.4模型(R^(2)=0.876 4,RMSE=0.033 0)精度提高最为显著,比DSR和SSR的R^(2)分别提高0.155 4和0.130 8且RMSE降低0.026 3和0.027 3。【结论】SCWC比DSR和SSR对针叶含水率更敏感,同时FP-SPA能快速准确地提取敏感光谱特征,因此SCWC对于针叶含水率的估算具有可靠的精度,从而可以实现落叶松毛虫危害下落叶松针叶含水率的准确预测。

【Objective】The early stress of insect infestation will cause small changes in the internal biochemical components such as the moisture content of larch needles,and the spectrum has a significant response.With the development of remote sensing technology,hyperspectral remote sensing can capture these small changes,and has become an important tool for monitoring and identifying early stress of insect pests.Accurate prediction of coniferous moisture content can provide an experimental basis for the early stress and early warning of Dendrolimus superans insect pests.【Method】Taking the forest insect pest outbreak area of the Daxing’anling as the study area,the hyperspectral data and coniferous moisture content data in July 2021 were obtained,and the spectral dimensionality reduction and extraction of sensitive features were obtained through three spectral features:smooth spectral reflectance(SSR),differential spectral reflectance(DSR)and spectral continuous wavelet coefficient(SCWC),combined with Findpeaks function(FP)and successive projection algorithm(SPA),and partial least squares regression(PLSR),random forest(RF)and support vector machine regression(SVMR)to establish a water content estimation model.【Result】1)Most of the sensitive bands of SCWC were concentrated in the three scales of 2^(2),2^(3) and 2^(4),and they were more sensitive than SR and SSR.2)FP-SPA quickly and accurately selected sensitive spectral features,reduced data redundancy,and reduced model complexity.3)The PLSR-SCWC-coif4 model had the highest accuracy(R^(2)=0.8904,RMSE=0.0375),which was 0.0963 and 0.1153 higher than the R^(2) of DSR and SSR,and 0.0075 and 0.0178 lower RMSE,respectively,followed by the SVMR-SCWC-bior4.4 model(R^(2)=0.8764,RMSE=0.0330)had the most significant improvement in accuracy.The accuracy was improved by 0.1554 and 0.1308 and decreased by 0.0263 and 0.0273 compared with DSR and SSR,respectively.【Conclusion】SCWC is more sensitive to coniferous moisture content than DSR and SSR,and FP-SPA can quickly and accurately extract sensitive spectral features,so SCWC has reliable accuracy in estimating coniferous moisture content,so as to accurately predict the moisture content of D.superans.