应用微型近红外光谱仪快速检测桑园土壤有机质含量的方法

Rapid Determination of Soil Organic Matter in Mulberry Field by Micro Near Infrared Spectrometer

采用低功耗、便捷式微型近红外光谱仪建立土壤有机质(soil organic matter,SOM)含量的快速检测方法,对于桑园土壤肥力监测和精确施肥有重要意义。对采集的71份桑园土壤样品,使用便携式NIR光谱仪采集样品的NIR漫反射光谱,使用一阶求导(1stD)、标准正态变量变换(SNV)和均值中心化(MC)3种方法预处理光谱,应用偏最小二乘法(PLS)建立预测模型,并用竞争性自适应重加权法(CARS)和随机蛙跳法(Random Frog)2种方法优选特征波长、优化模型,提高模型的预测精度。结果表明,使用1stD、SNV和MC预处理组合方式效果最好,2种波长优选方法均可提高模型预测精度,其中Random Frog-PLS方法效果最佳,建模集均方根误差(RMSEC)和相关系数(RC)分别为0.46%和0.94,交叉验证集均方根误差(RMSECV)和相关系数(RCV)分别为0.62%和0.89。对37个未知桑园土壤样品进行验证,RMSEP和相关系数RP分别为0.64%和0.90,实测值与预测值的误差较小,相关性较高,表明模型预测能力较好。研究表明,将便携式微型近红外光谱仪用于SOM含量的快速分析,可以促进对桑园土壤肥力的高效管理。

The development of rapid detection method for soil organic matter( SOM) with portable low-power micro nearinfrared( NIR) spectrometer is of great significance for soil fertility monitoring and precise fertilization in mulberry filed. A total of 71 soil samples from mulberry fields were collected. After the diffuse reflection NIR spectra were recorded,the spectra were pretreated by the 1st Derivative( 1stD),Standard Normal Variate( SNV) and Mean Center( MC) methods.Then,the Partial Least Squares( PLS) method was employed to establish the prediction model. Furthermore,Competitive Adaptive Reweighted Sampling Method( CARS) and Random Frog method were used to optimize the characteristic wavelength to improve prediction accuracy of the model. The results showed that the combination of 1stD,SNV and MC is the best pretreatment method and the two optimization methods can improve the accuracy of model.Among them,Random Frog-PLS has the best effect. The mean square root error of the calibration( RMSEC) and the correlation coefficient( RC) are 0. 46% and 0. 94 respectively,and the mean square root error of cross validation( RMSECV) and the correlation coefficient( RCV)are 0. 62% and 0. 89 respectively. Afterwards a prediction set of 37 unknown samples were verified, the meansquare root error of prediction set( RMSEP) and the correlation coefficient( RP) are 0. 64% and 0. 90 respectively. The error is small and the correlation is high between the measured and predicted values,which indicated that the model has good prediction capability. This study showed that the rapid determination of SOM content with portable Micro NIR spectrometer is feasible,which can promote the efficient management of soil fertility in mulberry field.