Rapid Identification of the Freshness of Pork Based on Near Infrared Spectroscopy and Back Propagation Neural Network Algorithm

Abstract

The freshness of pork is a critical factor in determining its quality and value, impacting its economic worth and suitability for consumption. Near-infrared spectroscopy was employed to collect the spectra of pork from day 1 to 9 at 18 ± 2°C. The backpropagation neural network (BPNN) algorithm, in conjunction with the previously mentioned near-infrared (NIR) spectral data, was utilised to construct a predictive model for the assessment of pork freshness, with the duration of storage serving as the primary variable. The model was evaluated in comparison to more traditional approaches, including partial least squares (PLS) and random forest (RF) models. The experimental results demonstrated that the BPNN model exhibited the most optimal test performance, with a determination coefficient (R²) of 0.93; the root mean square error of prediction (RMSEP) was 0.62 days. Furthermore, the mean absolute error of the test set (MAE) for the BPNN model was 0.48 days, indicating satisfactory prediction results. The experimental data demonstrated the feasibility of the proposed method in accurately estimating pork freshness, thus providing a novel technological reference for meat detection.