Proceedings Article | 5 December 2024
KEYWORDS: Data modeling, Polarization, Hyperspectral imaging, Education and training, Statistical modeling, Nondestructive evaluation, Cross validation, Convolutional neural networks, Statistical analysis, Modeling
In order to solve the problems of complex operation, destructive effect on the internal tissue of beef of traditional methods for detecting beef quality, the Convolutional Neural Network (CNN), based on near-infrared (900-1700nm) polarization hyperspectral multi-parameter imaging, is applied. The optimal prediction model of beef quality detection was established. The hyperspectral images of beef samples were collected under both unpolarized and polarized conditions, and the polarization hyperspectral data of the Region of Interest (ROI) were extracted, and multi-parameter values such as color (L* , a* , b*) and texture (hardness, adnesiveness, cohesiveness) of beef were measured. The Successive Projections Algorithm (SPA) was used to extract the characteristic wavelengths of the spectrum. Finally, the beef quality prediction model was constructed based on Convolutional Neural Network (CNN). Compared to the prediction results of unpolarization (Model 1), the polarization prediction model (Model 2) showed higher prediction accuracy, with a nearly 13% higher prediction accuracy advantage for parameter L*. The coefficients of determination in cross-validated (RCV2CV ) of L* , a* , b*, hardness, adnesiveness and cohesiveness in the polarization prediction Model were 0.84787, 0.70893, 0.94859, 0.6922, 0.67123 and 0.66254, respectively. The Root Mean Square Errors by Cross-Validated (RMSECV) were 0.98023, 0.91247, 0.4614, 3889.7131, 89.7464 and 0.032469, respectively. The feasibility of NIR polarization hyperspectral imaging technology in non-destructive detection of beef quality attributes was verified.
