Quantitative inversion model of protein and fat content in milk based on hyperspectral techniques

Traditional chemical methods for detecting milk composition suffer from many disadvantages, such as low efficiency and complicated operations. We propose a novel method based on hyperspectral inverse modelling method that combined Savitzky–Golay and first differentiation (SG_FD) to process the spectral data, coupled with an innovative application of improved spatial frog-hopping algorithm (IVRF_CA) to filter the feature wavebands, followed by a voting regressor (VR) to predict the fat and protein content in milk. The results demonstrated that the SG_FD algorithm is a hyperspectral preprocessing method that effectively improves the modelling accuracy, and the IVRF_CA algorithm reduced model complexity while ensuring the accuracy of the model. The test set coefficients of determination (R2) for the fat and protein partial least squares regression (PLSR) models built using feature wavebands filtered by the IVRF_CA were 0.9608 and 0.8623, respectively, while the corresponding test set R2 for the VR model were 0.9834 and 09607, respectively.