基于无人机多光谱遥感和机器学习的芳樟矮林长势反演

UAV Multispectral Remote Sensing and Machine Learning-based Growth Inversion for Cinnamomum camphora

摘要:

采用无人机多光谱技术有助于快速准确获取芳樟的长势信息，为芳樟矮林田间精准管理提供技术支持。本文以南方红壤区的芳樟矮林为研究对象，采用多光谱相机获取芳樟矮林冠层遥感影像，通过田间实测获取芳樟叶片叶绿素相对含量（SPAD）、叶面积指数（LAI）和地上生物量（AGB）数据，基于熵权法构建综合长势监测指标（CGMI），利用支持向量机（SVM）、反向传播神经网络（BPNN）、径向基函数神经网络（RBFNN）、决策树（DT）、多层感知机（MLP）、极端梯度提升（XGBoost）算法反演芳樟SPAD值、LAI、AGB、CGMI，对比分析单一指标和综合长势监测指标模型反演精度，最终选取最优模型。结果表明：基于6种机器学习算法分别构建芳樟矮林CGMI指标及单一指标反演模型，均表现出对芳樟CGMI反演效果最好，其模型测试集决定系数R2为0.614~0.862，均方根误差RMSE为0.074~0.953；在对CGMI指标的反演中，XGBoost模型在6种算法中的精度最高，其模型测试集决定系数R2为0.862，均方根误差RMSE为0.092。综上所述，对CGMI指标的反演可以准确判断出芳樟矮林的长势状况，XGBoost模型是芳樟矮林长势反演的最优模型，研究结果可为基于无人机多光谱技术的芳樟矮林长势监测提供参考依据。

Abstract:

UAV multispectral technology was demonstrated to facilitate rapid and accurate acquisition of growth parameters for Cinnamomum camphora (L.) presl var. linaloolifera Fujita, providing technical support for precision management of its dwarf forests. Cinnamomum camphora dwarf forests in the southern red soil region were investigated. Multispectral canopy remote sensing images were acquired by using a multispectral camera, while field measurements were conducted to obtain leaf chlorophyll content (SPAD), leaf area index (LAI), and above-ground biomass (AGB) data. A comprehensive growth monitoring index (CGMI) was then constructed through the entropy weight method. Six machine learning algorithms, support vector machine (SVM), back propagation neural network (BPNN), radial basis function neural network (RBFNN), decision tree (DT), multilayer perceptron (MLP), and extreme gradient boosting (XGBoost), were employed to invert SPAD, LAI, AGB, and CGMI values of Cinnamomum camphora. The inversion accuracy of single indices and the CGMI model was compared, and the optimal model was selected. The results demonstrated that all CGMI-based inversion models outperformed single-index models. The test set coefficient of determination (R2) ranged from 0.614 to 0.862, while the root mean square error (RMSE) ranged from 0.074 to 0.953. Among CGMI inversions, the XGBoost model achieved the highest accuracy (R2 was 0.862, RMSE was 0.092). In conclusion, CGMI inversion accurately assessed the growth status of Cinnamomum camphora dwarf forests, with XGBoost being the optimal model. The research result can provide a reference for UAV multispectral-based growth monitoring of such forests.