Hyperspectral inversion modeling of lactose content in milk based on CARS_VR

Achieving rapid and accurate determination of the components contained in milk is important for ensuring milk quality, preventing adulteration and safeguarding health. In this paper, we use hyperspectral technique to accurately measure the lactose content in milk. Herein we present a novel method based on hyperspectral inverse modeling utilizing Savitzky-Golay and first differentiation (SG_FD) to process the spectral data. The feature wavebands are filtered using competitive adaptive re-weighted sampling (CARS), and the lactose content in milk is predicted using an improved voting regression (VR). Upon comparing the predictions with other algorithms, it was determined that 39 feature wavebands screened with the CARS algorithm as input to the VR model had the best prediction effect, with a coefficient of determination $R_V^2$ of 0.9478 for the test set and a root mean square error RMSE
V
of 0.1382. It is evident from this paper that the improved VR inversion modeling algorithm presented in this paper has the potential to enhance both the overall performance of the inversion model and the prediction effectiveness of the model. The method can be applied to the detection of lactose content in milk and offers a new approach to determining the lactose content of milk.