Deep learning has been widely applied for Raman spectroscopy. This repository contributes PyTorch implementation of vanilla CNN and LSTM for Raman spectrum recognition (Liu et al., 2017; Yu et al., 2021). On this basis, this repository adopts two novel deep learning models: (1) graph neural networks (GNN) (2) contrastive learning, for Raman spectrum recognition.
The code has been tested running under Python 3.9.12, with the following packages and their dependencies installed:
numpy==1.16.5
pandas==1.4.2
pytorch==1.7.1
sklearn==0.21.3
python main.py --c 2 --model LSTMParameter model can be one of these four models: CNN, LSTM, GCN, CLR.
GNN is implemented as graph convolutional networks (GCN) (Kipf et al., 2017) with spectral angle for graph construction. Contrastive learning model is implemented as SimCLR with maximum mean discrepancy (Chen et al., 2020; Zhang et al., 2022).
This repo validates the models on two datasets proposed by Yu et al. (2021). The region of the Raman spectra data is from 600 to 1800 cm-1. The dimension of the Raman spectra data is 1200. Following Yu et al. (2021), the Raman spectra of each sample is preprocessed by 0-1 normalization.
The binary classification dataset bin.csv consists of the Raman spectra data from two kinds of microbes, Acinetobacter baumannii (Label 0) and Pseudomonas nitritireducens (Label 1).
The multi-class classification dataset multi.csv consists of the Raman spectra data from eight strains of Urechis unicinctus, named SX-1 to SX-8 (Label 0 to 7).
Deep learning models can achieve 99% accuracy on binary classification dataset, and 95% accuracy on multi-class classification dataset. For comparison, this repo also implements SVM and random forest for this task, see baseline.py.
We adopt an argument parser by package argparse in Python, and the options for running code are defined as follow:
parser = argparse.ArgumentParser()
parser.add_argument('--use-cuda', default=False,
help='CUDA training.')
parser.add_argument('--seed', type=int, default=1, help='Random seed.')
parser.add_argument('--epochs', type=int, default=200,
help='Number of epochs to train.')
parser.add_argument('--lr', type=float, default=0.01,
help='Learning rate.')
parser.add_argument('--wd', type=float, default=1e-5,
help='Weight decay (L2 loss on parameters).')
parser.add_argument('--hidden', type=int, default=64,
help='Dimension of representations')
parser.add_argument('--c', type=int, default=2,
help='Num of classes')
parser.add_argument('--d', type=int, default=1200,
help='Num of spectra dimension')
parser.add_argument('--model', type=str, default='CLR',
help='Model')
args = parser.parse_args()
args.cuda = args.use_cuda and torch.cuda.is_available()Chen et al. A simple framework for contrastive learning of visual representations. ICML. 2020
Deng et al. Scale-adaptive deep model for bacterial raman spectra identification. IEEE JBHI. 2022
Gu et al. Conformal prediction based on raman spectra for the classification of chinese liquors. Appl. Spectrosc. 2019
Ho et al. Rapid identification of pathogenic bacteria using Raman spectroscopy and deep learning. Nat. Commun. 2019
Kipf et al. Semi-supervised classification with graph convolutional networks. ICLR. 2017
Liu et al. Deep convolutional neural networks for raman spectrum recognition: a unified solution. Analyst. 2017
Liu et al. Dynamic spectrum matching with one-shot learning. Chemometr. Intel. Lab. Sys. 2019
Yu et al. Analysis of raman spectra by using deep learning methods in the identification of marine pathogens. Anal. Chem. 2021
Zhang et al. Divide and Contrast: Source-free Domain Adaptation via Adaptive Contrastive Learning. NeurIPS. 2022
Zhong et al. Accurate prediction of salmon storage time using improved raman spectroscopy. J. Food Eng. 2021