摘要
土壤有机质是衡量土壤质量与退化程度的重要指标.传统土壤农化分析方法具有耗时长、成本高等缺点,而高光谱技术具有的无损、快速特性表现出巨大的优势.采用两种特征变换方法对土壤光谱进行预处理,结合混合蛙跳算法(SFLA)进行光谱敏感波段选择,最后使用XGBoost算法建立土壤有机质含量反演模型.实验结果表明,基于SFLA-XGBoost模型的土壤有机质含量高光谱反演方法具有良好的应用潜力,使用原始光谱、连续统去除变换光谱、倒数对数光谱进行波段选择后得到的测试集数据R2分别为0.71、0.81和0.77.与偏最小二乘回归(PLSR)、高斯过程回归(GPR)两种模型进行对比,结果表明SFLA-XGBoost模型具有更高的精度与鲁棒性.
Soil organic matter is an important indicator of soil quality and degradation.Traditional soil agrochemical analysis methods have the disadvantages of time-consuming and high-cost,while the non-destructive and rapid characteristics of hyperspectral technology show great advantages.In this paper,two feature transformation methods are used to preprocess the soil spectrum,combined with Shuffled Frog Leaping Algorithm(SFLA)to select the spectral sensitive band.Finally,the XGBoost algorithm is used to invert the soil organic matter content.The results show that the hyperspectral inversion method based on SFLA-XGBoost model has excellent practical application potential,and the R~2 of the test set obtained by using the original spectrum,continuum removal transform spectrum and reciprocal logarithm spectrum for band selection are 0.71,0.81 and 0.77,respectively.Compared with Partial Least Squares Regression(PLSR)and Gaussian Process Regression(GPR),the results show that SFLA-XGBoost model has higher accuracy and robustness.
作者
杨素妨
YANG Sufang(Baise University,Baise 533000,Guangxi China)
出处
《河南科学》
2021年第5期720-728,共9页
Henan Science
基金
广西高校中青年教师基础能力提升项目(2020KY19021)。
关键词
土壤有机质
高光谱
混合蛙跳算法
特征选择
XGBoost算法
soil organic matter
hyperspectral
Shuffled Frog Leaping Algorithm
feature selection
XGBoost algorithm
