近红外光谱法对土壤有机碳组成的无损分析

Near infrared spectroscopy nondestructive analysis of soil on composition of organic carbon

利用近红外光谱(NIR)结合化学计量学方法建立一种定量监测土壤有机碳组成的新方法.对109个土壤样品进行近红外光谱采集,按常规方法测定土壤样品中总有机碳(SOC)及土壤样品中胡敏酸(HA)和富里酸(FA)有机碳的含量,再应用径向基函数(RBF)人工神经网络建立土壤样品的近红外光谱和有机碳组成含量间的数学校正模型.所建模型通过遗传算法(GA)自动选择最佳光谱处理方法和最优网络拓扑结构,同时构建了全光谱数据神经网络和GA自动优化光谱波段的神经网络模型,讨论了敏感光谱的选取对模型预报性能的影响.采用多元散射校正(MSC)光谱处理后建立的RBF神经网络模型为最优,该模型的预测精度和稳定性较好,可发展为复杂土壤系统中定量分析土壤有机碳组成的新方法.

Near infrared spectroscopy（NIR）combined with chemometrics method was used to develop a new method for the quantitative monitoring of soil on composition of organic carbon.The NIR spectra of the 109 soil samples were recorded and soil on composition of organic carbon were determined at the same time via the reference method.Radial basis function（RBF）neural networks was used to model the relationship between the NIR and the organic carbon concentration.The suitable spectral preprocessing method and network parameters were optimized automatically by using GA.The neural network models of full spectrum data and GA automatic optimization spectral band were constructed,and the influence of spectral band selection on model prediction performance was discussed.The RBF neural network model was optimized by multiplicative scatter correction（MSC）spectrum optimization.The results show that the predictive capability and the stability of this model were satisfied and this method should be popular in quantitative determination of soil on composition of organic carbon.