Analysis of Protein and Fat in Milk Using Multiwavelength Gradient-Boosted Regression Tree

Traditional chemical measurement methods for the milk composition are not only time-consuming and laborious but also highly polluting. This has necessitated the development of a new method to facilitate fast, easy, and real-time determination of milk composition. This article presents the use of a multichannel infrared spectral sensor and broadband infrared (IR) light source to obtain multi-wavelength feature data simultaneously. Furthermore, the gradient-boosted regression tree (GBRT) algorithm was used to develop a method for accurate milk content determination under different conditions. To this end, we developed a near-infrared (NIR) light-strength-acquisition device and accompanying software, compared the effectiveness of different machine learning algorithms, and established an optimal prediction model. Subsequently, the optimal prediction network was selected depending on the milk composition, thereby realizing the highest prediction accuracy. The results obtained in this study revealed that the milk protein and fat contents could be determined from the NIR absorption multispectra based on machine learning of the corresponding samples with coefficients of determination (R-2) values of 0.949 and 0.996, respectively. The corresponding root-mean-squared estimation errors of the prediction were 0.058 and 0.085, respectively. These experimental results indicate that the proposed milk quality evaluation system can be used to obtain real-time results. Moreover, it is simple, fast, affordable, and environmentally friendly.

Automated summary

This study presents a fast and accurate method for measuring milk composition using an infrared spectral sensor and machine learning algorithm. The results show that the proposed system can provide real-time, simple, and fast determination of milk protein and fat content.