Real-time detection of microbial aerobic plate count in salmon fillets based on near-infrared-hyperspectral imaging

Aerobic plate count (APC) detection in raw salmon fillets is an important guarantee of food safety. However, the existing flat colony counting method of APC detection is complex, time-consuming, and highly specialized, and is not suitable for on-site real-time detection. Near-infrared hyperspectral imaging (NIR-HSI) was used to establish quantification and exceedance discrimination models for APC. Salmon fillets exposed to air after cutting were sampled (99) and their APC was measured by flat colony counting. NIR-HSI detection of the samples was also performed. Average spectra of the effective region of interest (ROI) were calculated for NIR analysis; the training, prediction and independent external validation sets were rigorously divided. Partial least squares (PLS) and partial least squares-discriminant analysis (PLS-DA) served as the basic algorithms; standard normal variate (SNV) and Norris derivative filtering (NDF) were used for preprocessing; the combined moving window (MW) of forward-optimization and multi-wavelength step-by-step phase-out (MWSP) of backward-optimization was used for wavelength selection. For quantification, the number of wavelengths (N) in the optimal MW-MWSP-PLS model was 31; for exceedance discrimination, the N in the optimal MW-MWSP-PLS-DA model was 15. In independent external validation, the root mean square error of quantification was 0.682, and the total recognition-accuracy rates of exceedance discrimination reached 96.0%. The results show that NIR-HSI is feasible for APC quantification and precise exceedance discrimination. This method is non-invasive, rapid, and simple, and can be used for on-site real-time detection. The proposed few - wavelength combination model can be used for the development of a small-scale dedicated spectral imaging analyzer.