Song Hai-yan, Qin Gang
Guang Pu Xue Yu Guang Pu Fen Xi. 2015 Dec;35(12):3360-3.
Calibration transfer of near infrared spectroscopy model is a key problem in the field of near infrared spectroscopy research. In this study, calibration transfer of near infrared spectroscopy model for soil organic matter prediction by using Finite Impulse Response (FIR) was analyzed. The specific research work and conclusions were as follows: Firstly, 59 soil samples were collected and detected by using ASD Fieldspec3 in different times. 50 soil samples called "master soil samples" were detected at the same time in 2012, and the other 9 soil samples called "target soil samples" were detected at the same time in 2013. Secondly, 41 soil samples as calibration samples were randomly selected from the "master soil samples", and soil organic matter prediction model was built by using partial least square (PLS) analysis. The other 9 "master soil samples" were predicted. The result shows that the prediction correlation coefficient is 0. 961, Root Mean Standard Error of Prediction(RMSEP) is 0.600%, and Standard Error of Prediction(SEP) is 0.597%. It indicates that it is feasible to predict "master soil samples" by using above PLS model. Then, 9 "target soil samples" were predicted by using above PLS model. The result shows that it is infeasible to predict "target soil samples" by using above PLS model. Finally, FIR was applied to realize calibration transfer. The result shows that, when the window size was 516, higher prediction accuracy was obtained. The prediction correlation coefficient is 0.706, RMSEP is 0.662%, and SEP is 0.430%. It indicates that it is feasible to realize calibration transfer by using FIR.
近红外光谱模型的校准转移是近红外光谱研究领域的一个关键问题。本研究分析了利用有限脉冲响应（FIR）进行土壤有机质预测的近红外光谱模型的校准转移。具体研究工作及结论如下：首先，采集了59个土壤样本，并在不同时间使用ASD Fieldspec3进行检测。2012年同时检测了50个称为“主土壤样本”的土壤样本，2013年同时检测了另外9个称为“目标土壤样本”的土壤样本。其次，从“主土壤样本”中随机选取41个土壤样本作为校准样本，采用偏最小二乘法（PLS）分析建立土壤有机质预测模型，并对另外9个“主土壤样本”进行预测。结果表明，预测相关系数为0.961，预测均方根误差（RMSEP）为0.600%，预测标准误差（SEP）为0.597%。这表明利用上述PLS模型预测“主土壤样本”是可行的。然后，用上述PLS模型对9个“目标土壤样本”进行预测。结果表明，用上述PLS模型预测“目标土壤样本”是不可行的。最后，应用FIR实现校准转移。结果表明，当窗口大小为516时，获得了较高的预测精度。预测相关系数为0.706，RMSEP为0.662%，SEP为0.430%。这表明利用FIR实现校准转移是可行的。
