Traditional milk quality evaluation methods are labor-intensive and lack real-time applicability. In this study, a rapid and non-destructive method was developed to assess milk spoilage, integrating near-infrared (NIR) spectroscopy and silver nanoparticle (AgNP) chromogenic techniques. The NIR-based PLSR models effectively predicted six key spoilage indicators, with \(\:{\text{R}}_{\text{P}}^{\text{2}}\) values exceeding 0.90 and satisfactory RMSEP. AgNP-based colorimetric analysis enabled lactic acid concentration prediction, with the BPNN model achieving \(\:{\text{R}}_{\text{P}}^{\text{2}}\) = 0.95612 and RMSEP = 1.92557, while spoilage classification models based on SVM and RF yielded accuracies of 92.9% and 93.6%, respectively. Multi-information fusion further improved performance. Feature-level fusion (using SFLA-selected NIR features) achieved \(\:{\text{R}}_{\text{P}}^{\text{2}}\) = 0.98922 and RMSEP = 2.16126, and decision-level fusion achieved 98.4% classification accuracy and reduced the misclassification rate for pre-spoiled samples to 2.9%. These results demonstrate the potential of this integrated method for accurate, efficient, and real-time monitoring of milk spoilage in modern dairy processing.
