应用DS算法消除室内几何测试条件对土壤高光谱数据波动性的影响  

Using Direct Standardization algorithm to eliminate the effect of laboratory geometric parameters on soil hyperspectral data fluctuate characteristic

作　　者：洪永胜[1,2] 于雷[1,2] 耿雷[1,2] 张薇[1,2] 聂艳[1,2] 周勇[1,2] 

机构地区：[1]华中师范大学地理过程分析与模拟湖北省重点实验室,武汉430079 [2]华中师范大学城市与环境科学学院,武汉430079

出　　处：《华中师范大学学报（自然科学版）》2016年第2期303-308,共6页Journal of Central China Normal University：Natural Sciences

基　　金：国家自然科学基金项目(41401232;41271537);中央高校基本科研业务费专项资金项目(CCNU15A05006;CCNU15ZD001)

摘　　要：土壤高光谱遥感是土壤近地传感器(Proximal Soil Sensing)研究的重要方向,具有方便快捷、无破坏、成本低等优点,可以高效地分析和估算土壤属性参数.土壤高光谱数据的采集在室内较野外容易控制环境因素(如土壤水分、土壤表面属性等),获取数据更加稳定且具有可重复性,因而,基于室内高光谱数据反演土壤属性参数在国内外已形成较为成熟的理论.但是,室内土壤光谱数据的采集方法缺乏统一的标准体系,限制了光谱数据的共享,关键的几何测试条件不同而引起的光谱差异也未能消除,导致不同几何测试条件观测的光谱数据所建模型的传递效果较差.该研究以江汉平原公安县的36个土壤样本为研究载体,通过在室内环境下设置光源入射角度(A)、光源到土壤表面距离(L)、探头到土壤表面距离(H)3个几何测试参数的不同梯度组合,采用ASD FS3地物光谱仪获取27个组合的光谱数据,利用偏最小二乘回归(Partial Least Squares Regression,PLSR)建立27个参数组合的土壤有机质含量(soil organic matter content,SOMC)的反演模型,分析不同参数组合对土壤高光谱数据离散性的影响,确定1组理想的几何测试参数组合用于其他26组参数的模型传递研究,探讨直接标准化(Direct Standardization,DS)算法在消除其他26组参数组合光谱差异方面的可行性.研究表明,A30L50H15是室内较为理想的土壤高光谱的几何测试参数组合;L对土壤光谱反射率的影响没有明显的规律,而土壤光谱反射率随A增大而增大、随H增大而降低;经过DS算法校正后的其他26个参数模型的验证RPD值均增加到5.54,基本无光谱信息的丢失,模型稳定,有效的解决了不同几何测试参数之间光谱数据的差异性问题.Soil hyperspectral remote sensing has become an important research area of proximal soil sensing,which is considered to be a fast,non-destructive and low-cost method for effectively analyzing the soil key parameters.Several studies have proposed that the environmental factors,such as soil moisture and surface attributes,are more stable to be controlled in the laboratory than that in the field,data collected in the laboratory environment are more stable and repeatable,making it widely approved by researchers.However,the data sharing based on laboratory environment was limited due to lacking of uniform geometric parameters standards on data acquisition and the transfer effect was not satisfied between models established by different geometric parameters because of disparity induced by distinct key parameters.In this paper,36 soil samples at0~20cm depth were collected as experimental material from Gong＇an County in Jianghan Plain.The beam angle（A）,lamp distance（L）,sensor distance（H）were combined at different levels in the laboratory conditions.Then 27 sets of soil spectral data were measured by an ASD FieldSpec3 instrument,and 27 sets of quantitative inversion models for soil organic matter content（SOMC）were built using Partial Least Squares Regression（PLSR）method.Meanwhile,the fluctuations of soil hyperspectral data caused by different geometric parameters were analyzed.An optimal dataset was chosen for research on data transfer between the other 26 models.The feasibility was discussed upon the generalization capacity of Direct Standardization（DS）algorithm between the optimal geometric parameter and other 26 ones.Results showed that the optimal geometric parameter of beam angle,lamp distance and sensor distance is 30°,50 cm and 15 cm,respectively.Lamp distance（L）has a great impact on soil spectral reflectance without a regular pattern.In contrast,the soil spectral reflectance increases as beam angle（A）enlarging,while decreases as sensor distance（H）elongating.The validation PRD

关 键 词：土壤有机质 高光谱 室内几何测试参数 模型传递 DS算法 光谱校正 

分 类 号：S127[农业科学—农业基础科学] TP79[自动化与计算机技术—检测技术与自动化装置]