Pet food brand discrimination via near-infrared spectroscopy combined with modified SPA-SI information enhancement

Abstract

For pet food customers, it is essential to identify authentic brands quickly, simply, and cost-effectively. In this study, 353 pet food samples from four brands were selected, and high-precision brand discrimination was achieved using near-infrared (NIR) spectroscopy. At the same time, a method of information enhancement (IE) is proposed to enhance the information obtained from nine characteristic wavelengths selected by the successive projections algorithm (SPA), and the random forest (RF) discrimination model is established. The results show that the m-SPA-SI₃-RF strategy proposed by us has the best model performance, with a total accuracy of 94.32% and a single accuracy of 88–100%, which is far better than the discrimination performance of Full-RF (total: 92.05%) and SPA-RF (total: 89.77%). The number of wavelengths was reduced from 183 to 9, a 95.08% decrease, providing a basis for developing spectral equipment for discrimination between low-cost pet food brands. This study not only achieved spectral discrimination of pet food brands for the first time but also provided a new method for pet food brand discrimination, expanded the application potential of spectral analysis technology in pet food detection, and laid the foundation for the development of low-cost, specialized spectral equipment.