The Research On The Spectroscopy Method Of Detecting Crop Growth And Its Sensor Development

In our country agricultural cultivation,the excessive and improper use of chemical fertilizers and pesticides has brought increasing serious environmental pollution and food safety problems.Our goal is to understand the growth status of crops,and apply fertilizer properly.Based on the theory of principle of spectroscopy,the prediction models of the growth of sugarcane were set up and the equipment to detect its growth was developed.Sugarcane(ROC22)was studied in this paper,the accuracy of several modeling methods in predicting chlorophyll content by using the leaf reflectance spectra was explored,including principal component analysis,simple linear regression,multiple linear regression,BP neural network,etc.The 5 principal components were extracted from reflection spectrum in visible and near infrared range,which were used to set up the multiple linear regression model(Model 1)and the BP neural network model(Model 2).The R2 between the predicted and measured values were respective 0.7852 and 0.8929.In addition,the models of taking the sensitive band in 731 nm and 785 nm as well as spectral index as input were also set up,taking into account the engineering application.And the R2 of the simple linear regression model based on NDVI(Model 3),the multiple linear regression model based on 2 sensitive bands and their RVI and NDVI(Model 4),and the BP neural network model(Model 5)were respective 0.7470,0.7691 and 0.8482.The results proved that the accuracy of BP neural network model was higher than that of multiple linear regression model.And all of the models,Model 1 had the highest prediction accuracy,with R2 0.8929.However,Model 1 based on the whole wave band spectral information limited its engineering application.And Model 5 with R2 0.8482 which just need 2 sensitive bands had a slightly lower precision than Model 1,which made it have greater engineering application value.Beyond that,a set of crop nutrition diagnosis instrument with low cost was developed,based on the sensitive bands.That set of equipment and ASD FieldSpec4 were used to measure the reflectance to different color cards through calibrating experiment.The measurement results between them showed a significant correlation,and R2 was 0.9267.That reflected the fact that the instrument had a very high accuracy,and possessed non-destructive,real-time and accurate diagnosis potential to crops.