Research On Soil Total Nitrogen Detection Method Based On Nirs

Nitrogen is an essential element for crops the growth and development.Mastering the dynamic changes of soil components,will be better for farmland management and monitoring crop growth,and help scientific fertilization to improve crop yields.The traditional method for determining soil total nitrogen is mainly related to chemical analysis.There many disadvantages in traditional method.For example,the process is complicated,time-consuming and laborious;generates chemical waste and pollutes the environment.The measurement period is too long to monitor the soil total nitrogen quickly,and which is difficult to meet the need for environmental protection and rapid monitoring.The development of chemometrics and computer technology has promoted the visible near-infrared spectroscopy analysis technology which has become an independent analysis technology.NIRS has many advantages,like fast analysis,high efficiency,non-destructive detection,no chemical reagents,no pollution,and high stability.so it is widely used in the field of rapid detection of soil nutrients.This subject will study determination of total nitrogen in soil with NIRS.The research area is farmland in the Chongzhou Chengdu.100 soil samples were collected,and the total nitrogen content in the samples was determined using the national standard method(Kjeldahl method),and the moisture content of the soil samples was measured in the laboratory.Then,the near-infrared spectra of the soil samples before and after drying are respectively measured.The main contents are divided into the following three parts:1.Spectral preprocessing:using moving average smoothing,Savitzky-Golsy smoothing,first derivative smoothing,second derivative smoothing,standard normal variable transformation(SNV),a total of 5 different spectral preprocessing algorithms to process soil sample spectra,and then combined with partial least squares to establish Quantitative regression prediction model.According to the corresponding model indicators,the experimental results of different spectral preprocessing algorithms are analyzed.2.Spectral modeling analysis:Comparative analysis of five spectral full-wavelength modeling algorithms to predict soil total nitrogen results,which are linear and non-linear,including multiple linear regression(MLR),principal component regression(PCR),partial least squares regression(PLSR),support vector regression(SVR),artificial neural network(ANN).And comparative analysis of the experimental results of competitive adaptive reweighted algorithm(CARS)and the randomfrog algorithm.3.The removing of soil moisture:In order to reduce the influence of soil moisture on spectrum modeling and analysis,Direct standardization(DS)algorithm is used to establish a conversion matrix F between the wet soil spectrum and the dry soil spectrum to eliminate moisture factors Interference.According to the results of spectrum modeling experiment:(1)Among the five spectral preprocessing algorithms,the SG smoothing spectral preprocessing algorithm has the best effect.(2)Compared with the five full-wavelength modeling algorithms,the experimental results show that PLSR is the optimal modeling algorithm,the decision coefficient~2=0.779,the root mean square error value is 55.97mg/kg,and the RPD value is 6.46.(3)In the Wavelength-selecting models algorithm,the experimental results show that CARS combined with PLSR modeling has the best effect.The determination coefficient~2=0.791,the value of RMSEP is 65.94mg/kg,the RPD value is 5.76,and the prediction result is good.Among them,10 wavelengths extracted by CARS algorithm are 1431nm,1442nm,1445nm,1477nm,1482nm,2650nm,2758nm,2838nm,2858nm,2870nm,2915nm and 2920nm.The experimental results show that the full-wavelength PLSR algorithm experimental results are slightly better than the characteristic wavelength CARS+PLSR algorithm.However,full-spectrum data has high dimensions,redundant information,and high model complexity.CARS combined with PLSR modeling algorithm removes redundant variables by extracting variables,which is better in model simplification.(4)Use the DS algorithm to process the soil moisture spectrum,and then use the SG algorithm for spectral pretreatment and full-wavelength PLSR algorithm to modeling.The results show,the determination coefficient~2=0.702,the value of RMSEP is200.89mg/kg,RPD=1.83.It means that the DS algorithm has the potential to remove the interference of soil moisture factors,but the root mean square error value is large,The accurate quantitative regression prediction model of soil total nitrogen cannot be well established,so further optimization and improvement are needed.