Rapid classification of tea using laser-induced breakdown spectroscopy coupled with a BP neural network optimized by improved SSA

Abstract

Laser-induced breakdown spectroscopy (LIBS) is a swift and potent analytical method utilized for element detection, owing to its distinct advantages in online/in situ detection. However, the analysis of LIBS spectra encounters a significant challenge during the acquisition phase due to measurement uncertainty caused by matrix effects and self-absorption. To improve the performance of LIBS analysis, LIBS combined with a deep learning method based on a back propagation (BP) neural network optimized by the improved sparrow search algorithm (ISSA) was proposed here for the recognition of seven different types of tea. In order to realize rapid green identification of tea, a total of 1050 spectral datasets from seven tea samples were obtained by laser-induced breakdown spectroscopy. The spectral datasets were preprocessed to eliminate noise and background interference, and eight principal components were extracted by principal component analysis (PCA). In order to solve the issues of slow convergence of the traditional BP neural network model and its tendency to fall into the local optimal value, a hybrid strategy incorporating a tent chaotic map, adaptive T-distribution variation, number of producers and dynamic adjustment of search space were introduced to improve the SSA, and then the BP neural network hyperparameters were optimized with the ISSA. Subsequently, the hyperparameters of the BP neural network were optimized using the ISSA. Finally, the model is compared with K-Nearest Neighbors (KNN), BP, SSA-BP and other models. The results show that the ISSA-BP model has the best performance, and the recognition accuracy is superior to other models, with a classification accuracy reaching 99.1%. In conclusion, LIBS combined with the ISSA-BP neural network model can quickly and accurately recognize tea types.