Deeply-recursive convolutional neural network for Raman spectra identification

Abstract

Raman spectroscopy has been widely used in various fields due to its unique and superior properties.

Classification of Cannabinoid Spectra Using Machine Learning

“…We thus argue that Cases 1 and 2 lead to drastically different accuracies because Case 2 contain a relatively small amount of spectra, in such a way that even a small increase in their number causes an important difference in the performances of the related CNNs. This is in accordance with other studies using thousands of spectra to train CNNs [51,69,70].…”

Machine Learning Identification of Organic Compounds Using Visible Light

“…Class imbalance has also been addressed in other molecular classification studies based on Raman spectral databases, which also tend to be heterogeneous with a data record distribution that over-represents compounds of particular interest in chemistry. − To address class imbalance, researchers have explored data-augmentation strategies such as peak shifting, noise addition, smoothing, spline interpolation, and polynomial reconstruction. These preprocessing strategies were implemented to reconstruct the data set before training deep learning classifier models.…”

“…We addressed this class imbalance issue by preprocessing the raw data before training the classifier using resampling and physics-based data-augmentation strategies analogous to those employed by other machine learning classifiers trained with Raman spectra. − By training the random forest model with preprocessed balanced data, we achieve molecular classification testing accuracies in the UV and visible regions better than 98%. Additional improvements can be expected with additional steps of model hyperparameter optimization.…”

Raman spectral classification algorithm of cephalosporin based on VGGNeXt

“…15 A deep recurrent convolutional neural network was later proposed, which achieved good prediction accuracy on several different open-source Raman spectroscopy databases and outperformed previous methods on the RRUFF dataset. 16 Raman spectra can be converted into image representations using continuous wavelet transform (CWT), and transfer learning techniques can be used to achieve accurate gasoline grade identification. 17 However, there are some issues with the current Raman spectroscopy deep learning classification methods: (1) the majority of the training and test datasets are generated by the same device, and the trained models may have problems with spectral model transfer between different instruments.…”

“…15 A deep recurrent convolutional neural network was later proposed, which achieved good prediction accuracy on several different open-source Raman spectroscopy databases and outperformed previous methods on the RRUFF dataset. 16 Raman spectra can be converted into image representations using continuous wavelet transform (CWT), and transfer learning techniques can be used to achieve accurate gasoline grade identification. 17…”