Structured-illumination reflectance imaging provides a new means for food quality detection. This original work investigated the capability of Structured-illumination reflectance imaging technique coupled with multivariate chemometrics to evaluate the microbial contamination in pork inoculated with Pseudomonas fluorescens and Brochothrix thermosphacta during storage at different temperatures. The prediction performances based on different spectrum and the textural features of direct component and amplitude component images demodulated from the SIRI pattern images, as well as their data fusion were comprehensively compared. Based on the full wavelength spectrum (420 – 700) nm of amplitude component images, the orthogonal signal correction coupled with support vector machine regression provided the best predictions of the number of Pseudomonas fluorescens and Brochothrix thermosphacta in pork, with the R_p² values of 0.870 and 0.906, respectively. Besides, the prediction models based on the amplitude component or direct component image textural features and the data fusion models using spectrum and textural features from direct component and amplitude component images cannot significantly improve their prediction accuracy. Consequently, structured-illumination reflectance imaging can be further considered as a potential technique for the rapid evaluation of microbial contaminations in pork meat.

In this work, one-step growth models using hyperspectral imaging (HSI) (400–1 000) nm were successfully developed in order to estimate the microbial loads, minimum growth temperature (T_min) and maximum specific growth rate (μ_max) of Brochothrix thermosphacta in chilled beef at isothermal temperatures (4–25) ℃. Three different methods were compared for model development, particularly using (Model I) the predicted microbial loads from partial least squares regression of the whole spectral variables; (Model II) the selected spectral variables related to microbial loads; and (Model III) the first principal scores of HSI spectra by principal component analysis. Consequently, Model I showed the best ability to predict the microbial loads of B. thermosphacta, with the coefficient of determination (R_v²) and root mean square error  in internal validation (RMSEV) of 0.921 and 0.498 (lg (CFU/g)). The T_min (–12.32 ℃) and μ_max can be well estimated with R² and RMSE of 0.971 and 0.276 (lg (CFU/g)), respectively. The upward trend of μ_max with temperature was similar to that of the plate count method. HSI technique thus can be used as a simple method for one-step growth simulation of B. thermosphacta in chilled beef during storage.