Identification of meat species by combined laser-induced breakdown and Raman spectroscopies

We study the effect of complementary spectral information based on combined LIBS (laser-induced breakdown spectroscopy) and Raman spectroscopy, including 3 options of LIBS, Raman and LIBS-Raman, on the improved classification accuracy of meat tissues of beef, mutton and pork. The BPNN (back propagation neural network) with input variables optimized by RF (random forest) was used to classify the 3 kinds of meat tissues. The model confusion matrix, Precision, Recall, Kappa, MAE (Mean absolute error), RMSE (Root mean square error) and other parameters were obtained by 10-fold cross-validation method to evaluate the 3 classification models, and the results of the three methods were compared. The results showed that the combined LIBS-Raman model has the highest classification accuracy of up to 99.42%, and superior to the other 2 separate methods in terms of model consistency and confidence degree, indicating that the combined LIBS-Raman method has significantly improved the recognition ability and classification accuracy of meat tissues, which took the advantage of utilizing the complementary spectral information obtained by both methods. Therefore, the combination of LIBS-Raman and BPNN is a fast and robust method for meat tissue identification. [Display omitted] •The combination of LIBS and Raman can realize the rapid and nondestructive identification of beef、mutton、pork samples at the atomic and molecular structure levels.•Random forest feature selection can help to extract valid features from the spectra data of meat samples and reduce the interference of invalid features.•By building and optimizing the BPNN model, the LIBS-Raman spectral data of three types of meat samples can be effectively identified with an accuracy of 99.42%. In addition, the BPNN model is proved to have good robustness through external data sets.