光谱重构油菜叶绿素含量快速检测方法及设备研制

为了实现油菜叶片中叶绿素含量的快速无损检测，开发了手持式多光谱成像系统用于采集油菜叶片在460，520，660，740，840和940 nm 六个波段的光谱图像。将一台能够采集可见光/近红外（380～1 023 nm）512个波段光谱图像但是价格高昂且体积大的室内高光谱成像系统作为参考仪器，将手持式多光谱成像系统作为目标仪器后，采用伪逆法（pseudo-inverse method）求得高光谱成像系统和多光谱成像系统两台仪器之间的转换矩阵F，从而实现6个波段的多光谱图像向512个波段的高光谱图像的重构，提高了手持式设备的光谱分辨率。运用偏最小二乘回归算法（PLSR）建立了重构的光谱与油菜叶片的叶绿素含量之间的关系模型。结果表明，重构的可见光范围内的光谱反射率与叶绿素浓度之间具有很强的相关性，PLSR回归模型建模集的决定系数R2_c为0.82，建模集均方根误差RMESC为1.98，预测集的决定系数R2_p为0.78，预测集均方根误差RMESP为1.50，RPD为2.14。虽然应用本文开发的手持式成像系统结合PLSR模型实现油菜叶绿素含量快速无损预测的精度低于基于室内高光谱成像系统获得的高光谱图像建立的PLSR模型（R2_c，RMESC，R2_p，RMESP和RPD分别为0.90，1.41，0.82，1.36和2.37），但是明显优于基于原始多光谱成像系统4个波段（460，520，660和740 nm）反射率建立的PLSR模型得到的结果（R2_c，RMESC，R2_p，RMESP和RPD分别为0.78，2.06，0.72，1.85和1.88）。表明光谱重构技术可提高多光谱成像预测油菜叶绿素含量的精度，并且与室内高光谱成像系统相比，开发的手持式设备具有体积小、成本低廉和操作简便等优点，可为田间油菜叶片的生理状态和养分检测及可视化表达提供技术支持。

Rapidly Detecting Chlorophyll Content in Oilseed Rape Based on Spectral Reconstruction and Its Device Development

In order to rapidly and nondestructively detect chlorophyll content in leaves,a handheld multi-spectral imaging system was developed in this study to collect multispectral images of oilseed rape leaves.The pseudo-inverse method was introduced to reconstruct the multispectral reflectance at 6 wavebands(460,520,660,740,840 and 940 nm)to the hyperspectral reflectance at 512 wavebands in the range of 379~1023 nm with the aim to improve the spectral resolution.The partial least square regression(PLSR)was then used to build a model to predict chlorophyll content in leaves based on the reconstructed hyperspectral reflectance.The results showed that the reflectance in the visible range of the reconstructed hyperspectral presented a high relationship with the chlorophyll content.The performance of PLSR model using reconstructed spectrum as inputs was evaluated using the parameters of the determination coefficient of prediction set(Rp^2),root mean square error of prediction(RMSEP)and residual prediction deviation(RPD)with the values of 0.78,1.50 and 2.14,respectively,which was better than that using original spectrum at 4 wavebands(460,520,660 and 740 nm)with the values of Rp^2,RMESP and RPD of 0.72,1.85 and 1.88,respectively.The results demonstrated that the combination multispectral imaging with spectral reconstruction technology could improve the predicting ability of the PLSR model and this technology can be used for monitoring physiology and nutrient status in oilseed rape leaves.