Near Infrared Spectroscopy Detection of Soybean Fatty Acid Content based on Neural Network combined with Genetic Algorithm

Abstract

In order to solve the problem of poor performance in rapid analysis of soybean fatty acid content using traditional methods, the author proposes a near-infrared spectroscopy detection method for soybean fatty acid content based on neural network combined with genetic algorithm. The author first collected sample spectra and preprocessed them using mean centralization and Savitzky Golay smoothing differentiation method. Then, the optimal band was selected through segmented combination modeling, and genetic algorithm was used to further screen wavelength points sensitive to content prediction modeling. Finally, the spectral data was decomposed using principal component analysis (PCA), and the score matrix was input into a 3-layer neural network for training. The optimal model was established through parameter optimization. The experimental results show that the calibration relative analysis error RPD of the model based on genetic algorithm combined with neural network is, indicating that the model has good prediction accuracy and stability. To further validate the model’s reliability in practical applications, the correlation coefficient R between the predicted values and the standard values within the model’s prediction range was found to be 0.986, indicating a strong linear relationship between them. The relative deviation of each value is less than 2%, and the standard deviation of the difference is 0.0091. Paired t-test =1.2706 was performed on both methods, which is less than 2.354. This indicates that there is no significant difference between the analysis results of this method and the standard method, verifying the strong predictive ability of the model in practical applications. The model designed by the author can accurately and efficiently complete near-infrared spectroscopy detection.