Prediction of Minced Pork Quality Attributes Using Visible and Near Infrared Reflectance Spectroscopy

The objective of the present study was to estimate minced pork meat quality using visible and near infrared (Vis-NIR) spectroscopy. Two hundred twenty five carcasses samples from longissimus dorsi muscle were scanned over the Vis-NIR spectral range from 350 to 1 015 nm and analysed for intramuscular fat (IMF), protein and moisture according to the official methods. Wavelet transform was employed to eliminate the spectra noise. Partial least square regression (PLSR) and support vector machine (SVM) were used to develop Vis-NIR spectroscopy models for chemical composition detection. According to calibration statistics, the best model to predict intramuscular fat content was developed by SVM with the denoised spectra, the correlation coefficient was 0. 889 for calibration and 0. 888 for validation. For protein and moisture, the best model was achieved with the PLS method with the correlation coefficient of 0. 869 and 0. 881 for protein calibration and validation sets and 0. 877 and 0. 848 for moisture calibration and validation sets, respectively. And all the ratios of standard deviation of validation set to root mean square error of prediction (RPD) were not more than 3. 0. Results indicated that it was possible to predict chemical composition in minced pork meat. As a fast predictor of meat quality using Vis-NIR spectroscopy, it is necessary to improve the precision and the robustness of the model for practice.