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

Bo Wang ¹ Junying Han ^1* Chengzhong Liu ¹ Jianping Zhang ² Yanni Qi ²

• ¹ Gansu Agricultural University, Lanzhou, Gansu Province, China
• ² Gansu Academy of Agricultural Sciences (CAAS), Lanzhou, Gansu Province, China

The protein content of flaxseed (Linum usitatissimum) is an important factor affecting its nutritional value and quality. Spectral technology combined with advanced modeling methods can achieve fast, accurate and low-cost protein content prediction. In this study, the visible light-near infrared hyperspectral imaging technology combined with fractional order ant colony optimization (FOACO) was used to determine the protein content of flaxseed. Firstly, 30 flaxseed varieties widely planted in northwest China were selected as research objects, the hyperspectral information and protein data of flaxseed were collected, and the sample set division based on the joint x-y distance algorithm was used to divide the calibration set and the prediction set after removing the outliers. Then, a partial least squares regression (PLSR) model based on the original spectra and the preprocessed spectra was established, and it was determined that the Savitzky-Golay (SG) smoothing method showed superior performance after processing. Compared with the traditional band selection method principal component analysis (PCA) and ant colony optimization (ACO), PLSR and multiple linear regression (MLR) models were established according to the wavelength selection results for spectral data modeling and prediction performance analysis. In the MLR model, the prediction accuracy of FOACO is 0.9248, root mean square error (RMSE) is 0.4346, relative prediction deviation (RPD) is 3.6458, and mean absolute error (MAE) is 0.3259. The results show that the MLR model based on FOACO (FOACO-MLR) shows significant advantages in the determination of protein content of flaxseed, especially in the prediction accuracy and stability of characteristic bands. By combining visible and near-infrared hyperspectral imaging system (VNIR-HIS) technology and FOACO wavelength selection algorithm, this study provides an efficient and rapid method for the determination of protein content of flaxseed, which provides reliable technical support for the accurate detection of nutritional components.