利用高光谱成像仪(光谱范围400～1 000 nm)对土壤含水率进行了无损检测。 比较了208个土样不同天数下土壤含水率与光谱变化、 不同质量含水量光谱的差异； 对比分析了不同光谱预处理方法、 不同方法提取特征波长、 采用多元线性回归（multiple linear regression， MLR）、 主成分回归（principal component regression， PCR）与偏最小二乘回归（partial least squares regression， PLSR）建模， 优选出最佳模型。 结果表明： 光谱曲线的反射率随着土壤含水率的增加而减小。 当超过田间持水率时， 光谱曲线的反射率会随着土壤含水率的增加而增大。 对比分析了不同预处理方法， 近红外波段优选出单位向量归一化预处理方法。 采用无信息变量消除法（UVE）、 竞争自适应加权采样（CARS）、 β系数法、 连续投影算法（SPA）方法提取特征波长为49， 30， 5和7。 为了减少数据冗余， 对UVE与CARS提取的特征波长进一步采用SPA方法进行特征提取， UVE+SPA， CARS+SPA提取特征波长数分别为5和8个。 在此基础上， 利用MLR， PCR和PLSR方法对400～1 000 nm范围的特征波长建立模型， 对比分析不同建模效果， 优选出β系数提取的特征波长的MLR模型。 最优的特征波长为411， 440， 622， 713和790 nm， 最优模型的预测相关系数Rp=0.979， 预测均方根误差RMSEP为0.763。 因此， 今后可采用不同波段对土壤含水率进行定量分析。

This article summarizes a near-infrared hyperspectral imaging technique was investigated for non-destructive determination of soil moisture content. A total of 208 soil samples were collected by hyperspectral imaging system. The differences of soil water content and spectral change, and the spectra of different water contents were compared. Different spectral preprocessing methods were analyzed and the characteristic wavelengths were extracted by different methods. MLR, PCR and PLSR modeling were used to optimize the best model. The results show that the reflectivity of the spectral curve decreases with the increase of soil water content, and the reflectivity of the spectral curve increases with the increase of soil moisture content when it increases beyond the field water holding capacity. With the increase of soil moisture content, the spectral reflectance of soil showed a decrease at first before increasing. When the soil moisture content is 30%, the reflectivity of soil spectrum increases. It is mainly because the soil moisture content exceeds the amount of soil surface water layer,form a double structure the soil can accommodate . The method of different pretreatment is analyzed, and the pretreatment method of normalization of unit vector is proposed. The number of characteristic wavelengths extracted by UVE, CARS, β coefficient, SPA were 49, 30, 5, 7, respectively. In order to reduce the data redundancy, the characteristic wavelengths of UVE and CARS were further extracted by SPA method. The number of characteristic wavelengths of UVE+SPA and CARS+SPA were 5, 8. On the basis of this, the MLS, PCR and PLSR methods were used to model the characteristic wavelengths of the range of 400~1 000 nm. The MLR model of the characteristic wavelengths extracted by β coefficient was obtained by comparing the different modeling results.The optimal characteristic wavelength is 411, 440, 622, 713, 790 nm. The prediction coefficient Rp=0.979 is the best model, and the RMSEP is 0.763.Therefore, the soil moisture content can be quantitatively analyzed in different bands in the future.