Research Of Vermiculite Matrix Quick-available Nitrogen Detection Equipment Based On Near Infrared Spectroscopy

It is a new attempt to adopt the method of precision agriculture for crop cultivation in desert facility agriculture.The development trend of desert facility agriculture is the combination of vermiculite matrix with low cost and high water and fertilizer retention.Available nitrogen is often used as a short-term rapid nitrogen supplement to promote crop growth and ensure crop yield.Accurate information on the content of available nitrogen in vermiculite matrix plays a guiding role in the utilization of water and fertilizer in desert protected agriculture.Taking vermiculite matrix in desert facility agriculture as the research object,the content of available nitrogen in vermiculite matrix was quantitatively detected by near-infrared spectroscopy,and the near-infrared spectroscopy prediction model and detection equipment of available nitrogen in vermiculite matrix with wavelength of 940～1660 nm were established.(1)The influence of different particle grades of vermiculite matrix samples on near infrared spectroscopy was revealed.114 samples of vermiculite matrix with three particle size grades and different moisture content were prepared.The spectral prediction model of matrix moisture content was established based on different vermiculite matrix particle size grades,and the interference of vermiculite based plasmid degree grade on spectral detection was discussed.By comparing and analyzing the prediction models after spectral data processing of original spectrum,multivariate scattering correction(MSC)and standard normal variable transformation(SNV)preprocessing algorithm,the optimization and de-interference effects of MSC and SNV on spectral data of different particle sizes of samples are studied.Among them,SNV preprocessing algorithm is the best for spectral data preprocessing optimization of 0.20～0.60 mm vermiculite matrix samples,partial least squares regression(PLSR)modeling prediction model,and the RPD of the model is 6.70.MSC and SNV have better preprocessing effect on the spectral data of smaller samples,that is,the smaller samples should be used as much as possible to establish a high-precision spectral prediction model.The prediction performance of the model can be improved by reducing the particle size difference of samples through physical pretreatment methods such as sample crushing and screening,and reducing the influence of spectral scattering on near-infrared spectroscopy detection combined with pretreatment algorithm.(2)The variation of vermiculite matrix samples with different moisture content on near infrared spectroscopy was studied.Based on 114 vermiculite matrix samples with different moisture content with particle size grade of 0.00～0.20 mm,the spectral prediction model of matrix moisture content is established to study the influence of vermiculite matrix moisture content on spectral detection.By comparing and analyzing the original spectral data and the preprocessed spectral data,as well as the spectral prediction model established by PLSR and multiple linear regression(MLR)after the extraction of characteristic variables by successive projection algorithm(SPA)and competitive adaptive reweighting algorithm(CARS),the best near-infrared spectral prediction model of vermiculite basic material moisture content is discussed.After eliminating the uncorrelated spectral data,the MLR modeling model of S-G smoothing preprocessing algorithm combined with spa feature extraction is the best,and the RPD is 11.75.(3)The effects of different available nitrogen contents in vermiculite matrix samples on near infrared spectroscopy were explored.Based on 144 vermiculite matrix samples with different available nitrogen content,a spectral prediction model of matrix available nitrogen was established to study the possibility of near-infrared spectroscopy detection of vermiculite matrix available nitrogen.By comparing and analyzing the original spectral data,the spectral data processed by various preprocessing algorithms,the original spectra after SPA,CARS and Si-PLS feature extraction,and the PLSR spectral prediction model of preprocessing combined with the spectral data after feature extraction,the best prediction model of available nitrogen content in vermiculite matrix is discussed.In the whole band spectral modeling,the prediction reliability of the model based on the joint algorithm of second derivative and S-G smoothing preprocessing is the highest,and the RPD is 12.14.After extracting feature variables,the prediction accuracy of the model is the best when the first derivative is combined with S-G smoothing preprocessing method,and the accuracy of CARS feature extraction prediction model is relatively the best,and the RPD of the best prediction model is 12.57.(4)The near infrared spectroscopy detection equipment and test platform of available nitrogen in vermiculite matrix were developed.The model parameters of key components such as near infrared spectrometer are determined through analysis;Based on the self-contained software of near-infrared spectrum and Matlab software,the collection,prediction and analysis software of near-infrared spectrum available nitrogen content is designed.Based on MFC,an intelligent integrated test platform of water and fertilizer based on available nitrogen content control is developed.The test results show that the relative detection error of the test sample is within 20 %,the correlation coefficient is 0.91,and the standard deviation is 140.99 mg/Kg.The performance of the developed near infrared spectroscopy detection equipment for available nitrogen in vermiculite matrix can meet the actual detection requirements.