Wavelength Selection Method for Near Infrared Spectroscopy Based on Iteratively Retains Informative Variables and Successive Projections Algorithm

As a wavelength selection algorithm, successive projection algorithm ( SPA) is used in the quantitative analysis of near infrared spectroscopy to simplify the model complexity and improve the prediction accuracy. According to the principle of SPA algorithm, SPA method can only ensure that there is low redundancy between the two wavelengths selected by two adjacent projections, but it does not guarantee that the selected variables must be effective variables , that is to say , the subset of variables screened by SPA may contain some uninformation variables or even interfering variables. We extract effective variables (i. e. strong and weak information variables) by iteratively retains informative variables ( IRIV ) , and then select wavelength by SPA based on effective variables , so as to solve the problem that the selected variables by SPA may contain uninformation and interfering variables , and improve the prediction accuracy of the model. The improved IRIV-SPA is applied to processing of two groups of public NIR spectral data. After the wavelengths are selected, the MLR model is established and compared with the FULL-PLSR model and some other commonly used high-perf4mance wavelength selection methods ( SPA, IRIV, RF ) to prove the effectiveness of the improved IRIV-SPA algorithm. The results show that the improved IRIV-SPA-MLR model has the best prediction accuracy under two opening test datasets , which greatly improves the prediction accuracy and simplifies the complexity of the model compared with other algorithms. The improved IRIV-SPA can achieve efficient dimensionality reduction and is an effective wavelength selection algorithm.

Automated summary

Successive projection algorithm (SPA) is used for wavelength selection in near infrared spectroscopy to simplify model complexity and improve prediction accuracy, but may result in subsets containing uninformation or interfering variables. Improved IRIV-SPA iteratively extracts effective variables and selects wavelengths based on them, achieving the best prediction accuracy in two test datasets.