激光与光电子学进展, 2020, 57 (9): 093002, 网络出版: 2020-05-06
耦合机器学习和机载高光谱数据的土壤含水量估算 下载: 1209次
Coupled Machine Learning and Unmanned Aerial Vehicle Based Hyperspectral Data for Soil moisture Content Estimation
土壤含水量 无人机 高光谱数据 机器学习 地理加权回归模型 soil moisture content unmanned aerial vehicle hyperspectral data machine learning geographical weighted regression model
摘要
准确估算土壤含水量(SMC),对干旱区的精准农业、水资源管理具有重要意义。针对传统估算方法和野外测量耗时、费力的问题,通过无人机平台获取新疆阜康市冬小麦样地的高光谱影像数据,分别利用一阶导数、二阶导数、吸光度、吸光度一阶导数(FDA)、吸光度二阶导数对原始高光谱数据进行预处理;采用随机森林(RF)、梯度提升回归树(GBRT)和极端梯度提升(XGBoost)三种算法进行特征变量重要性遴选,基于地理加权回归(GWR)建立模型。结果表明:FDA的预处理效果最佳,以FDA-GBRT为基础的模型效果最优,建模集与验证集的决定系数(R2)分别为0.890、0.891,四分位数间隔为3.490;GBRT算法相较于RF和XGBoost算法优势较为突出,多数模型建模集与验证集的R2均大于0.600;GWR模型对SMC的预测建模有效,可为干旱区农业生态系统的管理与保护提供理论支撑。
Abstract
Accurate estimation of soil moisture content (SMC) is of great significance for precision agriculture and water resources management in arid areas. Traditional estimation methods and field measurements are time consuming and labor intensive. Therefore, we obtain hyperspectral image data of winter wheat plots in Fukang City, Xinjiang by unmanned aerial vehicle platform, and the original hyperspectral data are preprocessed through first derivative, second derivative, absorbance, first derivative of absorbance (FDA), and second derivative of absorbance. Random forest (RF), gradient boosted regression tree (GBRT), and extreme gradient boost (XGBoost) are used to select the importance of feature variables. A model is established based on geographical weighted regression (GWR). The results show that the pretreatment effect of FDA is the best. The model based on FDA-GBRT is optimal. The determination coefficient (R2) of the modeling set and the verification set are 0.890 and 0.891, respectively, and the quartile interval reaches 3.490. Compared with RF and XGBoost algorithms, the advantages of the GBRT algorithm are more prominent. The R2 of most of the model modeling set and the verification set are greater than 0.600. This indicates that the GWR model is effective in predictive modeling of SMC and can provide theoretical support for the management and protection of agro ecosystem in arid regions.
田美玲, 葛翔宇, 丁建丽, 王敬哲, 张振华. 耦合机器学习和机载高光谱数据的土壤含水量估算[J]. 激光与光电子学进展, 2020, 57(9): 093002. Meiling Tian, Xiangyu Ge, Jianli Ding, Jingzhe Wang, Zhenhua Zhang. Coupled Machine Learning and Unmanned Aerial Vehicle Based Hyperspectral Data for Soil moisture Content Estimation[J]. Laser & Optoelectronics Progress, 2020, 57(9): 093002.