Common salts classification using Raman spectroscopy and deep learning

In this work, a Convolutional Neural Network (CNN) model is proposed for classifying common salts based on their Raman spectral data. The data set is taken from the RRUFF database, contains Raman spectra of different salt samples. To enhance model generalization, the data augmentation process is utilized and the model trained under two preprocessing pipelines: First one is noise reduction and smoothing, and second one is baseline correction with noise reduction and smoothing. Experimental evidence shows that pipeline 1 reported a greater classification accuracy of 99.31% with less test loss at 0.0194 than pipeline 2 that reported 95.00% accuracy with 0.1471 test loss. Although CNN-based classification from Raman spectra has been investigated in previous research, this research is new in concentrating specifically on the classification of common salts, an underexplored area with significant application in environmental and industrial domain. The results indicate that efficient preprocessing considerably improves model performance and that deep learning has considerable potential for the identification of salts through Raman spectroscopy. Subsequent work can investigate larger datasets from diverse sources and incorporate spectral variations to further enhance model robustness and generalization.