Prediction of Pork TVB-N Content and pH Using Broad Learning System Based on Hyperspectral Imaging with Hybrid Wavelength Selection

This study proposed a non-destructive and accurate method for the detection of pork freshness based on hyperspectral imaging (HSI) and broad learning system (BLS). BLS models were developed and evaluated for their ability to predict the total volatile basic nitrogen (TVB-N) content and pH in pork samples based on hyperspectral images. Four different preprocessing methods (Savitzky-Golay (SG) smoothing, normalization, baseline correction, and standard normal variate) were applied to optimize the spectral data, and feature extraction was performed using competitive adaptive reweighted sampling (CARS), successive projections algorithm (SPA), and interval variable iterative space shrinking approach (iVISSA). The results indicated that SG was the best preprocessing method, and combining iVISSA with SPA for feature extraction effectively removed redundant features and reduced interference from irrelevant information, achieving optimal prediction performance in the BLS regression models. Specifically, for TVB-N prediction, the iVISSA-SPA-BLS model exhibited excellent performance with correlation coefficient of prediction (R_P) of 0.9422, root mean square error of prediction (RMSEP) of 3.0072, and residual prediction deviation (RPD) of 2.8038. For pH prediction, the R_P, RMSEP and RPD were 0.8173, 0.3679, and 1.7164, respectively. The developed method not only enables efficient and non-destructive prediction of pork freshness, but also provides a new non-destructive approach for food safety detection.