Rapid quantitative detection of psychrotrophic bacteria in raw milk using near-infrared transmission spectroscopy

Dairy spoilage caused by psychrotrophic bacteria and their metabolically derived heat-resistant enzymes poses a significant challenge to the global dairy industry. However, conventional detection methods are often hampered by limitations that restrict their application for rapid microbial safety assessment. To tackle this issue, this study presents a quantitative detection approach founded on near-infrared transmission spectroscopy to detect psychrotrophic bacteria in raw milk. Near-infrared spectra of raw milk samples were acquired within the wavelength range of 900 to 1,700 nm using a near-infrared spectrometer. Spectral preprocessing was then performed using moving window smoothing, Savitzky-Golay smoothing, standard normal variates, and multiplicative scatter correction. Four feature selection methods, including interval variable iterative space shrinkage approach, competitive adaptive reweighted sampling, iterative retained information variable, and uninformative variable elimination, were employed to extract feature wavelengths. Prediction models for psychrotrophic bacteria were then developed using partial least squares regression, random forest, and long short-term memory network algorithms. Among these, the SG-UVE-LSTM model exhibited the best performance, achieving a determination coefficient for the prediction set of 0.9277, a root mean square error of prediction of 0.2820 log cfu/mL, and a residual prediction deviation of 3.8256. This study offers an efficient and reliable approach for monitoring raw milk contamination by psychrotrophic bacteria.