Recent Progresses in Machine Learning Assisted Raman Spectroscopy

摘要

With the development of Raman spectroscopy and the expansion of itsapplication domains, conventional methods for spectral data analysis havemanifested many limitations. Exploring new approaches to facilitate Ramanspectroscopy and analysis has become an area of intensifying focus forresearch. It has been demonstrated that machine learning techniques canmore efficiently extract valuable information from spectral data, creatingunprecedented opportunities for analytical science. This paper outlinestraditional and more recently developed statistical methods that arecommonly used in machine learning (ML) and ML-algorithms for differentRaman spectroscopy-based classification and recognition applications. Themethods include Principal Component Analysis, K-Nearest Neighbor, RandomForest, and Support Vector Machine, as well as neural network-based deeplearning algorithms such as Artificial Neural Networks, Convolutional NeuralNetworks, etc. The bulk of the review is dedicated to the research advances inmachine learning applied to Raman spectroscopy from several fields,including material science, biomedical applications, food science, and others,which reached impressive levels of analytical accuracy. The combination ofRaman spectroscopy and machine learning offers unprecedentedopportunities to achieve high throughput and fast identification in many ofthese application fields. The limitations of current studies are also discussedand perspectives on future research are provided.