基于近红外光谱土壤水分检测模型的适应性

由于土壤水分的近红外光谱定量分析模型精度依赖于样品状态，故土壤水分定量分析模型的适应性极其重要。以湖北地区的3种土壤为研究对象，利用偏最小二乘法交叉验证建立了处理后样品下的土壤水分分析模型，模型预测值与标准值的决定系数R2为0.9946，交叉验证预测均方差为0.801%，模型预测决定系数R2为0.9919，预测均方差为0.912%；利用主成分分析了未处理土壤样品与处理土壤样品得分图的差异，结果表明定量分析模型对未处理样品的预测精度降低；采用斜率/截距的方法修正了12个未处理样品的模型预测值，预测平均绝对值误差从0.78%降低到0.38%，结果表明斜率/截距校正法能较好的提高近红外光谱土壤水分定量分析模型的适应性。

Adaptability of the model for soil moisture measurement based on near-infrared spectroscopy

The precision of the soil moisture measurement using near-infrared spectral quantitative analysis model relies on the sample condition, so the model adaptability is extremely important. Three kinds of Hubei area soil were researched, and partial least square (PLS) and cross calibration method were employed to establish soil moisture analysis model. The results indicate that the decision coefficient R2 between predicted value by model and normal value was 0.9946, and the root mean square error of cross-validation (RMSECV) was 0.801%, the model predicted decision coefficient R2 was 0.9919, the root mean square error of prediction (RMSEP) was 0.912%. Principal component analysis(PCA) method was used to classify the raw soil samples and the processing soil samples. However, the results indicate that the quantitative analysis model has low prediction precision for raw sewage sample. After the slope /bias method was used to revise 12 raw sewage sample values predicted by the model, the average absolute error reduced from 0.78% to 0.38%. The results indicate that the method of slope/bias can enhance the adaptability of the model for near-infrared spectral quantitative analysis of soil moisture.