Quasi-Qualitative Determinative Method Of FT-NIR Spectroscopy And Its Application In Soil Detection

Near-infrared(NIR)spectroscopy is a rapid analysis technique.NIR spectroscopy analysis must achieve a quantitative analysis or qualitative analysis of unknown samples by mathematical and statistical measurement methods to establish calibration model.Nevertheless,due to the serious overlap of NIR spectral signals,there is no obvious peak that can reflect the information of a single component to be tested.The calibration models always performs too ideally well to believe when established by the routine linear analytical methods.That is not convinced for the practical application in on-line detection.In this paper,statistical and multi-model integrated modeling methods are used for a fault-tolerant mechanism to be plug-into the quantitative analytical model,transform the fourier transform near infrared(FT-NIR)quantitative mode into a quasi-qualitative discriminant mode.A new discriminant method was proposed for quasi-qualitative determination by combining the interval search principal component analysis algorithm with logistic regression(iPCA-LR)or support vector machine(iPCA-SVM).Based on the common quantitative predicted value,the fault-tolerant threshold was set as different values.The samples were marked as accurate or non-accurate discriminated according to the priori predictive values and the thresholds,so that the original quantitative calibration method was transformed into a new quasi-qualitative discriminant method.The iPCA-LR method and i PCA-SVM method were applied for the FT-NIR quasi-qualitative discrimination.In the same process,we also discussed the latent variable extraction based on different wavebands that were generated by tuning the waveband division number.Some informative FT-NIR wavebands were selected with optimal discriminant accuracy.And for each fixed value,the i PCA-LR model and i PCA-SVM model were set by using different optimal wavebands.FT-NIR quasi-qualitative discriminant model of soil total nitrogen and organic matter were established,respectively.The optimal partial least squares iPCA-LR(PLS-iPCA-LR)model of total nitrogen: the worst optimal accuracy climbed to the level slightly above 75%.Meanwhile,some combination of informative wavebands were also tested and re-modeling for the discriminant accuracy.And the test of different informative wavebands or the combination of informative wavebands output optimal models with the accuracy above 90%;the optimal partial least squares iPCA-SVM model(PLS-iPCA-SVM): the worst optimal accuracy climbed to the level slightly above 68%.And the test of different informative wavebands output optimal models with the accuracy above 86%;the optimal artificial neural networks iPCA-LR model(ANN-iPCA-LR): the worst optimal accuracy climbed to the level slightly above 80%.And the test of different informative wavebands output optimal models with the accuracy above 90%;the optimal artificial neural networks iPCA-SVM model(ANN-iPCA-SVM): the worst optimal accuracy climbed to the level slightly above 70%.And the And the accuracy of different optimal wavebands for each interval is 1.The optimal PLS-iPCA-LR model of organic matter: the worst optimal accuracy climbed to the level slightly above 74%.And the test of different informative wavebands output optimal models with the accuracy above 88%;the optimal PLS-iPCA-SVM model: the worst optimal accuracy climbed to the level slightly above 70%.And the test of different informative wavebands output optimal models with the accuracy above 75%;the optimal ANN-iPCA-LR model: the worst optimal accuracy climbed to the level slightly above 80%.And the test of different informative wavebands output optimal models with the accuracy above 95%;the optimal ANN-iPCA-SVM model: the worst optimal accuracy climbed to the level slightly above 73%.And the accuracy of different optimal wavebands for each interval is 1.Results show that the metrological method transforms the quantitative prediction problem into the quasi-qualitative discriminant problem and deals with the disadvantages of overfitting and overidealistic modeling that always appears in common quantitative analysis.The quasi-qualitative discriminant mode was beneficial for the prediction effect of statistical model,more suitable for actual cases in field detection.The quasi-qualitative measurement method of NIR spectral analysis is expected to provide a technical reference for the rapid and accurate identification of agricultural substance components.