Identification of Edible Gelatin Origins by Data Fusion of NIRS, Fluorescence Spectroscopy, and LIBS

The potential for a data fusion of near infrared spectroscopy (NIRS), fluorescence spectroscopy, and laser-induced breakdown spectroscopy (LIBS) was investigated to improve the identification accuracy of different origins of edible gelatin (porcine skin, porcine bone, bovine skin, bovine bone, and fish skin). Competitive adaptive reweighted sampling method (CARSM) was applied to extract feature variables, and the feature variables from individual spectroscopic methods were combined to form the fused data. Then, random forest model (RFM) was built for classification of five origins of edible gelatin. The classification accuracy in the validation set for individual spectroscopic methods and the data fusion strategy were obtained as 97.1%, 98.55%, 81.16%, and 100%, respectively. Moreover, the precision, recall, and F score for the data fusion method were all up to 100%, which are apparently higher than those for the individual spectroscopic methods. The results demonstrate that the data fusion of NIRS, fluorescence spectroscopy, and LIBS can complement each other and improve the accuracy for discrimination of gelatin origins.