Studies On Calibration Transfer Methods Of Rapid And Nondestructive Near-infrared Detection For Individual Crop Kernels

Near Infrared Spectroscopy(NIRS)is a rapid and non-destructive analysis technique.High throughput screening of individual crop kernels by NIRS has the potential to improve the quality of agricultural products,improve the efficiency of crop breeding,and promote the development of plant phenomics.To establish a good individual crop kernel NIR calibration model requires a lot of time,manpower and material resources.However,a variety of changes from instruments,measurement environments,and samples may affect the prediction results and render the model inapplicable.Therefore,the improvement of the accuracy and applicability of the multivariate calibration model under different conditions is an urgent problem to be solved in the application of NIRS.The calibration transfer can correct the above differences and standardize the model.This dissertation aims to develop new calibration transfer methods to increase the universality and accuracy of NIR models in the analysis of individual crop kernels,and to promote the further application and popularization of NIRS in cereal production and crop breeding.The main research achievements are as follows:1.Evaluation and analysis of calibration transfer between various spectral platforms based on slope bias correction(SBC).Two groups of samples with ideal morphology and composition distribution(urea granules and starch granules)and two groups of unprocessed natural state samples(two groups of single rice kernels)are taken as target sample sets.Based on a classic and widely used calibration transfer algorithm-SBC,the transfer of their NIR models between three spectral platforms with different characteristics is studied.The results indicate that SBC can accurately calibrate the transfer of ideal state sample models between NIR platforms,and the transfer results(the root mean square errors of prediction(RMSEP)are 2.86 to 15.4)are close to or even better than the modeling results of the original platforms(the root mean square errors of cross validation(RMSECV)are 1.49 to 10.8).However,the calibration effect of SBC on the transfer of single rice kernel models between NIR platforms is poor.Therefore,it is necessary to develop new transfer methods to correct the difference of physical and chemical states in the spectral analysis of individual crop kernels and to improve the calibration transfer efficiency.2.A correlation analysis based wavelength selection(CAWS)method is proposed to optimize the transfer of NIR quantitative models.Based on correlation analysis,the proposed method can evaluate and screen the wavelengths with consistent and stable responses between the two sets of instruments,so as to improve the transfer efficiency of quantitative models.The application of this method is introduced by analyzing the transfer of one publicly available corn data set and one truely tested rice bran data set between different instruments.The superiority of CAWS is verified by comparing the CAWS method with other wavelength selection methods under the condition of different pretreatment and calibration transfer algorithms.The results show that the optimal models of the two batches of data sets optimized by CAWS obtain lower RMSEP than those of the other wavelength selection methods under the same conditions.In particular,the RMSEP of CAWS optimized corn model in calibration transfer is 0.070,which is better than those of the same data set reported by other literatures(RMSEP were 0.1519 and 0.098,respectively).Therefore,the proposed method can effectively improve the accuracy of the quantitative model when it is transferred,and has the potential to be applied to the transfer of quantitative model of individual crop kernels.3.A method of wavelength selection based on CAWS to improve the transfer efficiency of qualitative model is proposed.The proposed method is optimized based on CAWS algorithm,in which the Matthews correlation coefficient(MCC)is used as the criterion for wavelength selection of qualitative model,so as to achieve the purpose of improving the transfer efficiency of qualitative models.By analyzing the spectra of a publicly available pharmaceutical tablet data set and two truly tested wheat kernel and corn kernel data sets,and comparing the transfer results under different treatment conditions,the superiority of the proposed method is verified.The MCC values of CAWS optimized tablet,wheat and corn models after the transfer in all conditions are the highest,the second highest and the highest(the MCC values respectively are 0.987,0.717 and 1),respectively.The results show that this method can effectively improve the transfer efficiency of individual crop kernel qualitative models.4.Transfer between the model of natural state samples and ideal state samples based on Spectral Space Transformation algorithm(SST).By transferring the spectra of a kind of natural state samples(single rice kernels,SRK)into the spectra of their glume interference removed natural state samples(single brown-rice kernels,SBK)or the spectra of their ideal state samples(rice flour,RF)with SST algorithm,and then using the corresponding models(SBK model or RF model)for prediction,the accuracy of NIR analysis results of the single rice kernel samples is improved.The predicted results of the SRK transferred SBK spectra and SRK transferred RF spectra are respectively close to(RMSEP=0.480)or superior to(RMSEP=0.401)the result of analyzing single rice kernels directly(RMSEP=0.423).The results indicate that the method can correct the interference of physical and chemical state differences on spectral analysis of single rice kernels to a certain extent,and thus improves the universality and accuracy of the models.The method has the potential to be applied to the NIR analysis of other individual crop kernels.