Feature extraction method of hyperspectral scattering images for prediction of total viable count in pork meat

This study aimed to investigate the capabilities of hyperspectral scattering imaging in tandem with Gaussian function, Exponential function and Lorentzian function for rapid and nondestructive determination of total viable count (TVC) in pork meat. Two batches of fresh pork meat was purchased from a local market and stored at 10°C for 1-9 d. Totally 60 samples were used, and several samples were taken out randomly for hyperspectral scattering imaging and conventional microbiological tests on each day of the experiments. The functions of Gaussian, Exponential and Lorentzian were employed to model the hyperspectral scattering profiles of pork meat, and good fitting results were obtained by all three functions between 455 nm and 1 000 nm. The Lorentzian function performed best for fitting the hyperspectral scattering profiles of pork meat compared with other functions. Both principal component regression (PCR) and partial least squares regression (PLSR) methods were performed to establish the prediction models. Among all the developed models, the models developed using parameters CE (scattering width parameter of Exponential function) and CL (scattering width parameter of Lorentzian function) by PLSR method gave superior results for predicting pork meat TVC, with RV and RMSEV of 0.92, 0.59 log CFU/g, and 0.91, 0.61 log CFU/g, respectively. In addition, based on the improved hyperspectral scattering system, parameter c which represented the scattering widths in all three functions gave more accurate prediction results, regardless of the modeling methods (PCR or PLSR). The obtained results demonstrated that hyperspectral scattering imaging combined with the presented data analysis algorithm can be a powerful tool for evaluating the microbial safety of meat in the future.