Remote Sensing Monitoring Of Winter Wheat Canopy Hyperspectral Remote Sensing Based On Different Pretreatment Methods

The traditional collection of crop information is time-consuming and laborious,and the acquired information lags behind the state of crop growth.Hyperspectral remote sensing can achieve real-time non-destructive access to crop information and provide decision support for precise crop management.Based on the hyperspectral information and agronomy parameters,this study investigated the response different of canopy spectral in winter wheat with field experiment consisting of four different varieties and three nitrogen levels and used techniques such as hyperspectral analysis techniques,multiple regression methods,and machine learning theory,an exploratory study was conducted on spectral pre-processing and modeling methods for real-time non-destructive monitoring of winter wheat growth to improve the accuracy of real-time nondestructive monitoring of winter wheat growth.The main conclusions are as follow:The canopy spectral reflectance of winter wheat in different varieties in the visible range,followed by the growth period showed the trend of decreasing first and then increasing,and finally rising to the highest during the grain filling period;in the near infrared range,it showed the trend of rising first and then decreasing,reached the highest in the heading period.The spectral reflectance of different canopy spectral of winter wheat under different nitrogen application levels showed no significant difference in visible light range;In the near-infrared range,the canopy spectral reflectance of Aikang58 and Yumai49-198 increased with the increase of nitrogen application rate,but Zhoumai27 and Xinong509 showed the trend of increasing first and then decreasing,and Spectral reflectance was highest in N15 conditional canopy.The leaf area index and leaf nitrogen content of winter wheat in different cultivars were the same with the growth period,and they all showed a trend of first increase and then decrease.Spectral data were preprocessed using move moving average method(MAM),first derivative(FD),and standard normalization variate transformation(SNV),and then partial least-squares regression(PLSR),support vector machine regression(SVMR),and BP neural network regression(BP)estimation model were separately built on the preprocessed spectral data,and meanwhile compared with the model established using the original spectral data.The results showed: the BP neural network regression models based on different spectral data were all superior to the other two models by the leaf area index or the leaf nitrogen content estimation models.The combination of first derivative(FD)and BP neural network regression(BP)performed best.The R2 of the training and validation set of the best estimation model of leaf area index were 0.864 and 0.852,respectively,and the RMSE were 0.871 and 0.905,respectively.The R2 of the training and validation set of the best estimation model of leaf nitrogen content were 0.876 and 0.847,respectively,and the RMSE were 0.05 and 0.055,respectively.