Hyperspectral Analysis Of Winter Wheat Seedlings Under Low Temperature Stress Of A Typical Cold Snap

Winter wheat is a major commodity food in China and low temperature frost damage has been a limiting factor in its safe production.Traditional physiological and biochemical indicators of wheat are mostly measured by destructive sampling in the field combined with chemical analysis in the laboratory,which is time-consuming,laborious and inefficient,and can only be limited to"point scale"sampling,not suitable for large areas.The use of hyperspectral remote sensing technology to obtain physiological indicators of winter wheat leaves provides data and technical support for rapid and non-destructive monitoring of crop growth,guiding wheat field management and promoting the development of precision agriculture.In this study,the content of chlorophyll（SPAD）and soluble sugar in wheat leaves was determined before and after cold wave and the hyperspectral reflectivity of wheat leaves was obtained.Based on different spectral parameters and pretreatment methods,the spectral characteristics and variation rules of winter wheat leaves before and after low temperature stress were discussed.Five characteristic variables with the highest correlation coefficient with chlorophyll content were selected by correlation analysis and the estimation model of soluble sugar content was constructed by combining principal component analysis with stochastic forest algorithm in machine learning.The results can provide technical support for large-scale rapid detection of winter wheat under low temperature stress.The main conclusions are as follows:（1）The overall trend of increasing SPAD and soluble sugar content in wheat leaves after stress compared with that before low temperature stress indicated that wheat would increase chlorophyll and soluble sugar content in the short term to enhance its stress resistance when it was stressed by low temperature.Some of the winter wheat leaves showed different degrees of chlorophyll and soluble sugar contents after the stress,which may be due to the differences in seedling growth and development between different varieties.（2）The spectral reflectance of wheat leaves after low temperature stress was significantly different from that before stress in some bands,specifically in the infrared regions of380～650 nm,710～860 nm and above 950 nm,the original spectral reflectance of wheat leaves after freezing was slightly lower than that of the control,and the spectral reflectance was lower in the visible region where chlorophyll absorption was better.The original spectral reflectance of wheat leaves after freezing was slightly lower than that of the control.（3）The accuracy of the two hybrid models constructed for the estimation of chlorophyll content of wheat leaves before and after low-temperature stress was low after cross-validation,indicating that the model for estimating chlorophyll content of wheat at room temperature was not applicable to the estimation of chlorophyll of wheat after low-temperature stress,and a separate model for estimating chlorophyll after low-temperature stress should be established;in the hybrid model for the inversion of chlorophyll content of winter wheat under low-temperature stress using spectral data,the first-order spectral derivative at The model constructed at 694 nm had the best estimation effect,with a goodness-of-fit（R²）of 0.694 and a root mean square error（RMSE）of 3.191,indicating that the method of using the characteristic spectral bands of wheat leaves to build an inverse model of leaf chlorophyll content under low-temperature stress is feasible,with the model expression Y=649.92(R′₆₉₄)^0.527.（4）For the construction of a hybrid estimation model for the soluble sugar content of wheat leaves,the number of feature bands extracted using PCA principal component analysis after four different pre-processing methods were:75 for the first-order differential transform,12 for the multiple scattering correction,5 for the fast Fourier transform and 7 for the SG filter smoothing.The model estimation results show that the R² of the training set is above0.88 and the RMSE is below 0.1 for the models constructed by different pre-processing methods,while the validation set has a large difference compared with the training set.Through screening,the spectral first-order derivative transformation-based method was identified as the model for estimating soluble sugar content in winter wheat leaves,and its regression equation was:Y=0.6478x+0.0221.