为了研究宁夏地区土壤的水分迁移机理以及对土壤水分快速无损检测, 利用高光谱成像(光谱范围900~1 700 nm)技术对土壤的含水率进行了研究。通过高光谱成像系统采集了208个土样, 比较了不同天数下土壤含水率与光谱的变化、不同质量含水量光谱的差异。对采集到的土样进行PLSR模型建立, 对比分析不同光谱预处理方法、不同方法提取特征波长(UVE、CARS、β系数、SPA)、不同建模方法(MLR、PCR、PLSR)建立的模型, 优选出最佳模型。结果表明: 在一定的土壤含水量范围内, 光谱曲线的反射率与土壤含水率成反比; 当增大到超过田间持水率时, 光谱曲线的反射率与土壤含水率成正比。对比分析了不同预处理方法, 优选出单位向量归一化预处理方法。对比不同的模型, 优选出SPA提取的特征波长的MLR模型。最优的特征波长为987, 1 386, 1 466, 1 568, 1 636, 1 645 nm, 最优模型的预测相关系数Rp=0.984, 预测均方根误差RMSEP为0.631。因此, 今后可采用不同波段对土壤含水率进行定量分析。

By using the near-infrared(spectral range of 900-1 700 nm) hyperspectral imaging technique, the soil moisture movement mechanism and non-destructive determination of the moisture content of soil in Ningxia Hui Autonomous Region were studied. A total of 208 soil samples were collected by hyperspectral imaging system. The differences among soil water content, spectral changes, and spectra of different water contents were compared. The best model was chosen by different spectral pretreatment manners, different extraction manners of the characteristic wavelengths(UVE, CARS, β coefficient, SPA), and different building model manners(MLR, PCR, PLSR). The results show that the changes of spectra are inversely proportional to the changes of soil water content within the moisture limits. When the moisture content of soil is beyond the field water holding capacity, the changes of spectra are directly proportional to the changes of soil water content. The unit vector normalized preprocessing method is optimized. The best model is MLR method based on the characteristic wavelength of SPA extraction. The optimal characteristic wavelengths are 987, 1 386, 1 466, 1 568, 1 636, 1 645 nm, and the value of optimal correlation coefficient and RMSE are 0.984 and 0.631, respectively. Therefore, the soil moisture content can be quantitatively analyzed using different wavelengths.