A unique cuvette and near-infrared spectral imaging for fast and accurate quantification of milk compositions

Abstract

Analyzing the nutritional compositions of milk using traditional methods is costly, time-consuming, and impractical for rapid field measurements, resulting in delays in quality control procedures. This paper presents a rapid, accurate, and easy-to-use method for quantitative analysis of finished milk compositions using a low-cost, portable, and environmentally friendly measurement system. The system uses a near-infrared (NIR) broadband digital camera to capture spectral images of milk after it is placed in a customized cuvette in the short-wave NIR region (700–1050 nm). A machine learning algorithm that combines image edge detection and gradient-boosted decision trees then regresses these images to predict the protein and fat content of the milk. Each measurement requires a maximum of 1.75 mL of milk sample with no additional consumables and takes 0.1 s to complete. The coefficient of determination (R²_CV) for the fat detection model was 0.999 and the root mean square error (RMSE_CV) was 0.026 g/100 mL. For protein detection, the R²_CV was 0.957 and the RMSE_CV was 0.058 g/100 mL. The experimental results show that the system achieves high precision and stability while realizing miniaturization and portability.