Flaxseed protein content prediction based on hyperspectral wavelength selection with fractional order ant colony optimization.

The protein content of flaxseed (Linum usitatissimum) is a crucial factor influencing its nutritional value and quality. Spectral technology combined with advanced modeling methods offers a fast, accurate, and cost-effective approach for predicting protein content. In this study, visible-near infrared hyperspectral imaging (VNIR-HIS) technology was combined with fractional order ant colony optimization (FOACO) to determine the protein content of flaxseed. Thirty flaxseed varieties commonly cultivated in Northwest China were selected, and hyperspectral data along with protein content measurements were collected. A joint x-y distance algorithm was applied to divide the dataset into calibration and prediction sets after removing outliers. Partial least squares regression (PLSR) models were developed based on both raw and preprocessed spectra, with the Savitzky-Golay (SG) smoothing method found to provide superior performance. The performance of wavelength selection methods based on FOACO, principal component analysis (PCA), and ant colony optimization (ACO) was compared using PLSR and multiple linear regression (MLR) models. The FOACO-MLR model achieved a prediction accuracy of 0.9248, a root mean square error (RMSE) of 0.4346, a relative prediction deviation (RPD) of 3.6458, and a mean absolute error (MAE) of 0.3259. The results show that the FOACO-MLR model provides significant advantages in predicting flaxseed protein content, particularly in terms of prediction accuracy and stability of characteristic bands. By combining VNIR-HIS technology with the FOACO wavelength selection algorithm, this study offers an efficient and rapid method for determining the protein content of flaxseed, providing reliable technical support for the precise detection of nutritional components.