Studying On Hyperspectral Estimation Model Of Biochemical Parameters Of Rape Under Indoor Nitrogen Stress

The relationship between the amount of nitrogen applied to crops and the content of spectra and biochemical parameters is the basis for the use of hyperspectral technology for crop nutrition monitoring.In this study,ground-based spectroscopy was used as a support,and the relationship between nitrogen fertilizer,biochemical parameters,and spectra was studied and analyzed.The optimum fertilization amount and harvest period for indoor rape planting in autumn and winter were determined,and a spectral estimation model for biochemical parameters was established.In this way,with the help of remote sensing technology,crop nutrition indicators can be quickly and easily,non-destructively,effectively and accurately monitored,and fertilization and dynamic regulation can be instructed to provide theoretical basis for providing scientific identification technology for vegetable food safety and the production of pollution-free vegetables.In order to obtain biochemical parameters and leaf spectral data sets of rape under nitrogen stress,a four-month pot experiment was conducted in the laboratory.Four nitrogen-graded experimental groups and one blank control group were set up in the experiment.The biochemical parameters(nitrogen content,chlorophyll,soluble protein,soluble sugar,nitrate,VC,dry matter,and water content)and leaf spectral data of the six growth stages were collected in a 20-day growth period.By analyzing the biochemical parameters of rapeseed during the whole growth period,it was found that the chlorophyll,soluble protein,soluble sugar,dry matter and water content all showed a trend of increasing first and then decreasing with the extension of growth period and nitrogen application rate.Nitrate was positively correlated with the amount of nitrogen applied,and its content decreased slightly with the growth of the leaves.The VC content decreased first and then increased with the delay of the growth period.The content of the VC was the lowest in the fast growing period and then increased but still did not exceed the seedling period.In conclusion,the best performance of different biochemical parameters was mostly in the mature stage of N2 experimental group.Therefore,taking into account the economic benefits and time cost,it is initially determined that the optimum fertilization amount for low-fertile soil indoor rape growing in autumn and winter season is 0.2g/kg.It is recommended that the best harvest period is 81~100 days after emergence,and the top-dressing period is rapid growth.Spectral studies on different nitrogen application rates at the same time showed that the reflectance in the visible light band decreased with the increase of nitrogen application rate,while the near infrared emission showed a slight instability with the increase of nitrogen application rate.Comparison of the amplitudes shows that the visible light band is more sensitive to changes in the amount of nitrogen applied,and changes in the near-infrared amplitude are not significant.This shows that nitrogen fertilizer mainly affects the spectral reflectance curve by changing the pigment content of the leaf,and has little effect on the structure of the leaf.The analysis of the correlation between the first derivative of the original spectrum and the spectrum and the biochemical parameters shows that there is a certain correlation between the two in the whole wave range,and the correlation coefficient shows a distinct peak-to-valley feature.The correlation between different biochemical parameters and spectral parameters is different.Chlorophyll,protein,soluble sugar and dry matter showed significant negative correlations with the visible and short-wave infrared reflectances of the original spectrum,and there was no significant positive correlation between the near-infrared bands.The correlation between the VC content and the original spectral reflectance in the visible and near-infrared bands was not significant,and there was a very significant positive correlation between the short-wave infrared1530 to 1858 nm and 2144 to 2301 nm.There is a significant negative correlation between nitrate and the original spectrum in the visible range,a very significant positive correlation in the near-infrared band,and a negative correlation in the short-wave infrared range.The water content showed no significant positive correlation in visible light,and negative correlation in near infrared and short wave infrared.It can be seen that different nitrogen application rates have different degrees of influence on the biochemical parameters and spectral reflectance of the leaf,and the biochemical parameters have a certain correlation with the spectral reflectance,which provides abundant data for further quantitative analysis of the relationship between spectral reflectance and biochemical parameters.Sensitive spectral parameters for different biochemical parameters are different.Through correlation analysis,spectral parameters with correlation coefficient greater than 0.6 were selected for modeling,including curve statistical model,stepwise multiple regression model,partial least squares regression(PLS)model,random forest algorithm and PROSPECT.The results show that the estimation ability of the five models is quite different,and the curve statistical model performs the worst.The modeling accuracy of this model is generally low and it is impossible to accurately estimate the content of each parameter of the blade.The accuracy of modeling based on stepwise multiple regression increased to between 0.642 and 0.802,but the model has great instability and low verification accuracy.The verification accuracy of the PLS model for each biochemical parameter is relatively stable.However,the ability to estimate dry matter and nitrate is not good.The estimating model based on the random forest algorithm has significantly improved the estimation effect of all parameters,the modeling accuracy(R~2)is above 0.88,the verification accuracy is above 0.91,the RE is between 0.047%and 2.429%,and the RRMSE is at 0.150%.7.115%.This shows that the random forest algorithm is superior to all the above models in estimating biochemical parameters.At the same time,the PROSPECT model was built based on the measured data,the spectral reflectance curve under the best N value was simulated,and the chlorophyll content was inverted using the cost function.The content of three biochemical parameters was indirectly estimated based on the linear relationship between the measured chlorophyll content and leaf nitrogen content,protein content and soluble sugar content.Thus,a quantitative relationship between the spectrum-chlorophyll-nitrogen content/protein/soluble sugar was initially established.