| Fertilizer is an important factor in agricultural production.Developing countries can increase total food yield by more than 30%through fertilization.As a kind of fertilizer with the largest scale of modern production and use,chemical fertilizer has the characteristics of high nutrient content and fast fertilizer effect.The nutrient elements in modern fertilizers mainly include a large amount of nitrogen,phosphorus,potassium,and other medium and trace elements.Although there are corresponding implementation standards for the nutrient content of chemical fertilizers,due to poor production materials,backward manufacturing processes,and expired deterioration,there are a large number of fertilizer products on the market with substandard nutrient content,which makes it difficult to apply chemical fertilizers accurately and poses a certain threat to crop growth and soil safety.In this study,33 kinds of chemical fertilizers commonly used in the existing market were taken as the research objects.The N,P2O5 and K2O contents of chemical fertilizers were determined by the current national standard method.Visible/near-infrared spectroscopy,hyperspectral imaging and LIBS were used to collect the spectral signals of chemical fertilizer under the original particle,solid powder and solid tablet.The regression model of N,P2O5 and K2O content was established using the spectral signals,and the multi-source spectral fertilizer quality analysis software was built based on it.The main research contents are as follows:(1)Based on the comparison of visible/near infrared spectroscopy and hyperspectroscopy,the detection model of N,P2O5 and K2O content in different forms of chemical fertilizers was established.A small visible light spectrometer and a near-infrared spectrometer were combined with a three-in-one optical fiber to collect the visible/near-infrared spectrum signals of the fertilizer samples in the state of powder and pressed tablets.The original spectrum curve showed that the signal of the dual spectrometer has good continuity.Combining the SG smoothing and SNV spectrum preprocessing methods,and using the CARS method to extract 1/10 of the full spectrum as the characteristic wavelength,the PLS and ELM models of the fertilizer powder and pressed tablets were established at 550nm～950nm,1050nm-1640nm,full spectrum and characteristic wavelengths.The results show that the modeling effect of chemical fertilizer in powder state is significantly better than that of pressed tablets.The modeling effect of 1050nm-1640nm near infrared band is better than that of 550nm～950nm visible light band,and the modeling effect of characteristic wavelength is close to that of full spectrum modeling.In the regression model based on fertilizer powder,the optimal Rp2 of N,P2O5,and K2O are 0.989,0.963,and 0.981,respectively,and the corresponding RMSEP values are 0.910,1.724,and 1.393.The near-infrared hyperspectroscopy was used to collect the near-infrared hyperspectral image of the fertilizer in the original particle state,and the regional average spectrum was extracted from the hyperspectral image,and the regression model of PLS,ELM and SVM was established.The optimal Rp2 of N,P2O5,and K2O are 0.987,0.994,and 0.988,respectively,and the corresponding RMSEP values are 1.009,0.986,and 1.084.(2)The LIBS collection process was optimized and the content detection models of fertilizer N,P2O5,K2O based on the LIBS spectrum were established.The spectrum collection process of the LIBS system under multi-sample and multi-collection points was optimized,to realize rapid positioning during multi-sample collection and the synchronization problem in the continuous multi-point collection process was solved.The LIBS path planning software was compiled to realize the automatic planning and control command generation of complex LIBS laser scanning path,which can simulate the LIBS collection path in advance,simplifying the spectrum collection operation.The optimized LIBS system was used to collect the spectral signal of the fertilizer in the tablet state,and the principal component analysis of the spectral signal was performed.The interpretation rates of the first three principal components PC1,PC2,and PC3 to the overall spectrum are 0.309,0.253,0.205,respectively.Using the coefficients of the first three principal components,4430 wavelength variables were selected as characteristic spectra.PLS and SVM regression models were established with the full spectrum of 22063 wavelength variables and characteristic spectra.The optimal Rp2 of N,P2O5,and K2O are 0.990,0.992,and 0.993,respectively,and the corresponding RMSEP values are 1.297,0.866 and 0.868.The model based on the characteristic wavelength and the full-spectrum model have close prediction accuracy.(3)Integration of portable VIS/NIR spectroscopy system and multi-source spectral fertilizer quality analysis software.That a small visible light spectrometer and a near infrared spectrometer was combined,realized the synchronous information collection and analysis of multiple spectral devices,and completed the portability of the whole system.A software system for real-time analysis of fertilizer quality that can be connected to a variety of spectrometers has been written.The software adopts a modular design,decouples the spectrum detection equipment and the spectrum analysis method,and realizes the rapid combination of the spectrum equipment and analysis method.It uses dynamic loading of functional modules to achieve flexible enhancement and iterative upgrades of software functions.The local software and cloud server were combined to realize online upgrades of spectroscopy equipment and analysis methods,and at the same time support reliable spectroscopy data upload to the cloud,and continuously improve the redundancy of fertilizer quality detection. |
PDF Full Text Request