Energy value measurement of milk powder using laser-induced breakdown spectroscopy (LIBS) combined with long short-term memory (LSTM)

Abstract

Milk powder can provide the necessary nutrients for the growth of infants, and the level of its energy value is an important factor in the measurement of its nutritional value. Therefore, the measurement of the energy value in milk powder is of great significance for the nutritional health of infants. In this study, samples of 32 different brands of milk powder were selected for spectral analysis, and laser-induced breakdown spectroscopy (LIBS) combined with deep belief network (DBN), back propagation (BP) neural network, and long short-term memory (LSTM) models was used to achieve quantitative analysis of the energy value of the milk powder. The experimental results revealed that the LSTM model outperformed the DBN and BP models in terms of accuracy, with a mean relative error (MRE_P) of 1.0029%, which was 73.03% lower than that of DBN (3.7186%) and 69.53% lower than that of BP (3.2914%). Moreover, the determination coefficient (R_P²) value improved significantly from 0.9341 for DBN and 0.9766 for BP to 0.9984. In addition, the root mean square error (RMSE_P) decreased to 0.2140 from 0.7042 for DBN and 0.9051 for BP. These results demonstrate that the LSTM model has superior predictive performance compared to the other models. Therefore, the combination of LIBS and LSTM can accurately measure the energy value of milk powder and provide an effective and feasible means for its commercial measurement.