Sensitive variables extraction, non-destructive detection and visualization of total viable count (TVC) and pH in vacuum packaged lamb using hyperspectral imaging

Abstract

The feasibility of hyperspectral imaging (HSI) for sensitive variables extraction, non-destructive detection and visualization of total viable count (TVC) and pH in lamb was investigated. Regions of interest (ROIs) of pure muscles were acquired and the corresponding representative dataset were preprocessed and divided into the calibration set and the prediction set using the method of sample set partitioning based on the joint X–Y distance (SPXY). Moreover, sensitive variables were studied and identified from full band spectra (473–1013 nm) by coarse screening (GA), medium screening (GA-CARS) and fine screening (GA-CARS-SPA). Consequently, sixty two and fifty nine feature-related variables were selected to develop the best models of GA-CARS-PLSR for predicting and visualizing TVC and pH in lamb with R²_p of 0.93 and 0.96 and RMSEP of 0.42 and 0.054, respectively. In addition, based on GA-CARS-SPA, seventeen and sixteen simplified feature-related variables were finally extracted to build new multiple linear regression (MLR) models for facilitating further development of a multispectral system of these two attributes, yielding R²_p of 0.79 and 0.87 and RMSEP of 0.73 and 0.11, respectively. The promising results indicate that HSI has the potential as a fast and non-invasive method for simultaneously predicting and visualizing multiple freshness attributes of vacuum packaged lamb samples.