基于高光谱和数字图像特征指数的受渍冬小麦SPAD估算

为了探索基于高光谱和数字图像技术的受渍冬小麦SPAD最优监测方法,本研究基于排灌可控的微区试验,通过分析常用的15个高光谱特征指数和14个数字图像特征指数与受渍冬小麦叶绿素相对含量(SPAD)的相关关系,构建了基于最优监测特征指数的BP神经网络模型,对受渍冬小麦的SPAD进行估算。结果表明: 与正常小麦相比,短期渍水(≤7 d)对冬小麦的SPAD值和高光谱反射率影响不明显,当渍水时间大于12 d时,SPAD值随着渍水时间的增加急剧降低,在生长后期接近于0;基于数字图像特征指数(绿红差值植被指数、超红指数、红光标准化值和超绿指数)的冬小麦SPAD估算结果,与基于相对应的高光谱波段的估算结果基本一致,估算模型实测值与预测值的R²最高达到0.86,均方根误差(RMSE)为3.98;与基于数字图像特征指数相比,基于类胡萝卜素反射指数、黄边幅值、归一化植被指数和结构不敏感指数4个高光谱特征指数的冬小麦SPAD估算模型的实测值与预测值的R²最高达到0.97,RMSE低至1.95。可见,基于高光谱和数字图像技术均可以进行受渍冬小麦SPAD估算,且基于高光谱特征指数的BP神经网络模型的估算结果较好。

Estimation of SPAD value in waterlogged winter wheat based on characteristic indices of hyperspectral and digital image

To explore the optimal monitoring method for soil and plant analyzer development(SPAD) of winter wheat under waterlogging stress based on hyperspectral and digital image techno-logy, the correlations between SPAD of the waterlogged winter wheat and fifteen indices of hyperspectral characteristic and fourteen indices of digital image feature were analyzed under a micro-plot which could be irrigated and drainage separately. Then, the BP neural network models for SPAD estimation were constructed based on the optimal monitoring feature indices. Compared with the normal winter wheat, SPAD and the value of hyperspectral reflectance did not change under short-term waterlogging(less than 7 d), whereas the SPAD was significantly decreased after more than 12 d waterlogging treatment with the value being close to zero at the late stage of growth. The estimation accuracy based on the digital image characteristics of green minus red, excess red index, norma-lized redness index and excess green index showed similar results compared to that using the BP network model based on the characteristics of the corresponding hyperspectral band. The highest R2 between the measured value and the predicted value was 0.86, while the root mean square error(RMSE) was 3.98. Compared with the BP network models built with the digital image feathers, the accuracy of the models based on the four hyperspectral characteristic indices(carotenoid reflex index, yellow edge amplitude, normalized difference vegetation index and structure insensitive pigment index) for SPAD was significantly improved, with the highest R2 of 0.97 and the lowest RMSE of 1.95. Our results suggest that both hyperspectral and digital image technology could be used to estimate SPAD value of waterlogged winter wheat and that the BP network model based on hyperspectral characteristic indices performed better in the estimation accuracy.