高光谱估算土壤有机质含量的波长变量筛选方法

发布时间：2018-07-20 12:03

【摘要】：土壤高光谱数据量大、波段维数高,存在光谱信息无效、冗余和重叠现象,导致基于全波段构建的土壤有机质含量反演模型不稳定、精度难以提升。因此,探寻筛选关键波长变量的方法,通过滤除干扰、冗余、共线信息,提高模型预测性能,是目前土壤高光谱研究的热点之一。该文对江汉平原公安县的土壤样本进行室内理化分析、光谱测量与处理等工作获取了实证数据,采用无信息变量消除法(uninformative variables elimination,UVE)剔除无效变量,利用竞争性自适应重加权算法(competitive adaptive reweighted sampling,CARS)滤除冗余变量,运用连续投影算法(successive projections algorithm,SPA)消除共线变量,并尝试将不同类型的筛选方法进行耦合筛选关键波长变量,应用偏最小二乘回归(partial least squares regression,PLSR)分别建立土壤有机质含量估算模型,对比各种变量筛选方法的优缺点,最终,构建筛选土壤高光谱数据关键变量的方法体系。研究结果表明,除SPA方法的模型精度低于全波段外,其他6种变量筛选方法的建模效果均优于全波段;在3种单个变量筛选方法中,CARS方法优于UVE、SPA变量筛选方法,能有效地筛选出重要波长变量,其预测集相对分析误差RPD值为2.84;综合比较各种变量筛选方法,发现CARS-SPA方法从全波段2 001个波长中筛选出37个特征波长建立的土壤有机质含量的PLSR模型效果最好,其模型预测集的决定系数R2和相对分析误差RPD值分别为0.92、3.60,所选波段仅为全波段的1.85%。CARS-SPA-PLSR模型简单、预测效果好,可作为该区域土壤有机质含量估测的重要方法,对今后土壤近地传感器设备的开发具有一定的指导作用。
[Abstract]:Because of the large amount of soil hyperspectral data, high wave band dimension, invalid spectral information, redundant and overlapping phenomena, the inversion model of soil organic matter content based on the whole wave band is unstable and the precision is difficult to improve. Therefore, it is one of the hotspots of soil hyperspectral research to explore the method of screening key wavelength variables and improve the performance of model prediction by filtering interference, redundancy and collinear information. In this paper, the indoor physical and chemical analysis, spectral measurement and treatment of soil samples in Jianghan Plain were carried out, and the invalid variables were eliminated by (uninformative variables elimination without information variable. The competitive adaptive reweighting algorithm (competitive adaptive reweighted sampling cars) is used to filter redundant variables, and the continuous projection algorithm (successive projections algorithm is used to eliminate collinear variables. The estimation model of soil organic matter content was established by partial least square regression (partial least squares regression), and the advantages and disadvantages of various methods were compared. Finally, the method system of selecting the key variables of soil hyperspectral data was constructed. The results show that the model accuracy of the SPA method is lower than that of the full-band method, and the modeling effect of the other six variable selection methods is better than that of the full-band method, and the car method is better than the UVE-SPA variable selection method among the three single variable selection methods. The relative analysis error (RPD) of the prediction set is 2.84. It was found that the PLSR model of soil organic matter content established by CARS-SPA method was the best, and 37 characteristic wavelengths were selected from 2 001 wavelengths. The determination coefficient R2 and the relative analysis error RPD of the model prediction set are 0.92n3.60, respectively. The selected band is only 1.85. CARS-SPA-PLSR model is simple, and the prediction effect is good. It can be used as an important method to estimate the soil organic matter content in this region. It can be used to guide the development of soil near-earth sensor equipment in the future.
【作者单位】：华中师范大学地理过程分析与模拟湖北省重点实验室;华中师范大学城市与环境科学学院;华中师范大学湖北经济与社会发展研究院;
【基金】：国家自然科学基金项目(41401232;41271534) 中央高校基本科研业务费专项资金项目(CCNU15A05006;CCNU15ZD001)
【分类号】：S153.621;S127

【相似文献】