大豆胁迫的快速无损检测对提高大豆品质和产量至关重要, 近年来, 通过高光谱技术进行植物胁迫的检测得到广泛应用, 但针对大豆水氮胁迫的应用研究鲜有报道。 针对开花结荚期大豆设置了4种水分和5种氮素水平进行胁迫实验, 获取高光谱、 叶绿素含量和净光合速率生理信息数据, 并通过光谱数据计算了15个光谱植被指数, 最终选取了5种植被指数, 分别为归一化差异植被指数NDVI、 比值植被指数RVI、 绿色归一化差异植被指数GNDVI、 改进红边归一化指数mNDVI705和叶绿素指数LCI以指示水氮胁迫对大豆的影响。 同时通过建立单叶叶绿素含量和净光合速率反演模型进行大豆生理信息的预测, 采用相关分析法提取敏感波段, 所提取的敏感波段分别为520~622和485~664 nm; 采用多元散射校正(MSC)、 标准正态变量变换(SNV)、 一阶导数(FD)、 二阶导数(SD)和Savitzky-Golay平滑(S-G)预处理方法, 同时采用主成分分析(PCR)和偏最小二乘(PLS)2种建模方法, 将其按一定关系组合成多种方法, 以相关系数为模型评价指标, 寻找出最优预处理与建模方法的组合。 结果表明: 未受胁迫和受胁迫大豆的高光谱曲线具有整体变化趋势一致但光谱反射率值不同的特征, 未受胁迫大豆的反射率在500~700 nm波段具有最低值, 在760~900 nm波段具有最高值; 随着水氮胁迫程度的增加, 500~700 nm波段的反射率逐渐增加。 不同水分和氮素水平对植被指数的影响不同, 但变化规律一致, 5种植被指数均表现为未受胁迫大豆大于受胁迫大豆, 且随着水氮胁迫程度的增加, 植被指数值逐渐减小。 建立反演模型所用最优方法组合为MSC+FD+S-G+PLS和SNV+SD+S-G+PLS, 校正集相关系数分别为0.960 6和0.992 7, 预测集相关系数分别为0.972 0和0.970 8, 表明所建模型的精度较高, 可对受胁迫和未受胁迫大豆单叶叶绿素含量和净光合速率生理信息进行精准预测, 为大面积种植时检测其生理信息提供技术支持。

Rapid and non-destructive testing of soybean stress environment are critical to improving soybean quality and yield. In recent years, the detection of plant stress by hyperspectral technology has been widely used, but there are few reports on the application of water and nitrogen stress in soybean. Four kinds of water and five kinds of nitrogen levels were set in the flowering and pod-forming soybeans for stress experiments in this paper. After the stress, the physiological information data of hyperspectral, chlorophyll content and net photosynthetic rate were obtained, and 15 spectral vegetations indices were calculated by spectral data. The index NDVI, RVI, GNDVI, mNDVI705 and LCI were used to indicate the effects of water and nitrogen stress on soybean. And soybean physiological information was predicted by establishing single leaf chlorophyll content and net photosynthetic rate inversion model. The sensitive region was extracted by correlation analysis, and they were 520~622 and 485~664 nm respectively. Multivariate scatter correction (MSC), standard normal variable transformation (SNV), first derivative (FD), second derivative (SD) and Savitzky-Golay smoothing (S-G) preprocessing are used, while two modeling methods, principal component regression (PCR) and partial least squares (PLS), are selected to combine them into several methods according to a certain relationship. The correlation coefficient is used as a model evaluation index to find a combination of optimal preprocessing and modeling methods. The results showed that the hyperspectral curves of non-stressed and stressed soybeans had the same trend but different spectral reflectance values. The reflectance of unstressed soybean has the lowest value in the 500~700 nm region and the highest value in the 760~900 nm region, and the reflectance in the 500~700 nm region gradually increases with the increase of the degree of water-nitrogen stress. The effects of different water and nitrogen levels on vegetation index were different, but the changes were consistent. The 5 vegetation indexes showed that the unstressed soybean was larger than the stressed soybean, and the vegetation index value decreased with the increase of the degree of water-nitrogen stress. The optimum combination of inversion models is MSC+FD+S-G+PLS and SNV+SD+S-G+PLS. The correlation coefficients of the correction set are 0.960 6 and 0.992 7, and the correlation coefficients of the prediction set are 0.972 0 and 0.970 8, respectively. The results show that the model has high precision, and can accurately predict the physiological information of chlorophyll content and net photosynthetic rate of stressed and unstressed soybean, and provide technical support for detecting physiological information during large-scale planting.