DeepRaman

A Universal and Accurate Method for Easily Component identification in Raman Spectroscopy Based on Deep Learning

Raman spectra contain abundant information from molecules but are difficult to analyze, especially for the mixtures. In this study, a method entitled DeepRaman has been proposed to solve this problems. Essentially, it is a pseudo-Siamese neural network (pSNN) with spatial pyramid pooling (SPP) to predict component(s) in an unknown sample by comparing their Raman spectra. DeepRaman obtains promising results for the analysis of surface-enhanced Raman spectroscopy (SERS) dataset of artificial sweeteners and Raman imaging dataset of gunpowder. Together, it is an accurate, universal and ready-to-use method for spectral matching and compound identification in various application scenarios.

Depends

pip install -r requirements.txt

Usage

1.Training your model

Run the file 'training.py'. Since the data exceeded the limit, we have uploaded some example data for training，download at releases.

2.Predict mixture spectra

Run the file 'testing.py'. Some data has been upload as examples. The input 1 and input 2 represent pure component spectrum and unknown spectrum, respectively. More example data for testing can be download at releases.

Contact

Xiaqiong Fan: fxq@haut.edu.cn Zhimin Zhang: zmzhang@csu.edu.cn