A general framework for qualitative analysis of Raman spectroscopy based on deep learning

Deep learning has become the prevailing method for qualitative analysis of Raman spectroscopy. However, for researchers and engineers without a background in computer science or artificial intelligence, the knowledge required can be challenging. This includes the underlying principles of deep learning and the features of frameworks such as PyTorch, TensorFlow, and Keras. To address this challenge, we summarize the Raman spectroscopy analysis process and design a suitable deep learning framework, including dataset creation, data preprocessing, data augmentation, deep learning model development, model training, model evaluation, model visualization analysis, and model deployment. We use four open-source Raman spectroscopy datasets, including binary classification and multi-classification examples, to explain the framework's components and operation and provide comprehensive evaluation results. This framework has the potential to improve work efficiency and accelerate the research process.