A cost-effective on-site milk analyzer based on multispectral sensor

• An analyzer on milk compositions was developed based on multispectral sensor. • Partial least squares regression was used to establish prediction models. • Homogenization can improve determination performance for fat and total solids. • The R V were 0.90 and 0.96 for fat content of raw and homogenized milk. • The portable and low-cost analyzer could be used in-situ detection. Monitoring the main compositions, i.e., fat, protein, lactose and total solids, of milk has become an indispensable part in cow management. However, long time and high cost in measurement make the traditional detection methods cannot realize rapid and on-site analysis. To overcome these shortcomings, a rapid and cost-effective milk analyzer based on a low-cost multispectral sensor was developed. Python language was used to develop the software of the analyzer. One hundred and fifty raw milk samples were used in this study. The diffuse reflectance of the raw milk and their homogenized milk were collected at 18 wavelengths by using the developed milk analyzer. Partial least squares regression was used to establish prediction models for each composition. The evaluation experiment on the accuracy of the developed analyzer showed that the root mean squares errors and correlation coefficients of validation set were 0.37% and 0.90, 0.34% and 0.57, 0.16% and 0.21, 0.53% and 0.86 for fat, protein, lactose and total solids content of raw milk and were 0.23% and 0.96, 0.31% and 0.63, 0.14% and 0.14, 0.41% and 0.91 for homogenized milk. The designed milk analyzer, just two hundred dollars in hardware cost and less than 1 kg in weight, makes on-site analysis on milk compositions possible.