110 / 2021-07-27 22:01:54

Feature Wavelength Selection in Near-Infrared Spectroscopy Based on Genetic Algorithm

Feature wavelength selection attempts to identify and remove the variables that penalize the performance of a model since they are useless, noisy and redundant or correlated by chance. Thus, feature wavelength selection has become a critical step in the modeling process of near infrared (NIR) spectral analysis. In this paper, genetic algorithm (GA) for feature wavelength selection is proposed. GA randomly simulates the natural selection of biological evolution to achieve global optimization and selects the optimal feature subset from the variable space. The root mean square error of cross-validation (RMSECV) is taken as the fitness function. Partial least squares (PLS) algorithm is used for regression analysis to establish the prediction model. Comparing with successive projections algorithm (SPA) and competitive adaptive reweighted sampling (CARS), the quantitative model based on GA has better prediction ability and robustness. The correlation coefficient of prediction (RP), the root mean square errors of prediction (RMSEP) and the residual predictive deviation (RPD) of GA-PLS model are 0.9906, 0.0482, 10.2929. The results show that the GA can reduce the number of modeling variables while ensuring the improvement of model prediction accuracy and robustness.