SemiRaman: A self-supervised contrastive representation learning-based framework for semi-supervised Raman spectral identification of pathogenic bacteria.

Raman spectroscopy has established itself as a robust and non-invasive approach for the precise identification of pathogenic bacteria. The fusion of Raman spectroscopy with advanced artificial intelligence algorithms offers a promising strategy for accelerating microbial detection, particularly in scenarios involving hazardous biological contamination and public health threats. However, the performance of current deep learning-based approaches is often hindered by the scarcity of labeled spectral data and the challenges associated with high-dimensional and noise contamination of Raman spectra. In this paper, we propose a concise and efficient self-supervised contrastive learning-based framework (SemiRaman) to achieve accurate semi-supervised Raman spectral identification of pathogenic bacteria with very limited labeled data. SemiRaman combines unsupervised and supervised learning modules, extracting discriminative features from unlabeled data through redundancy reduction and multi-level contrastive learning, and achieving high classification accuracy with minimal labeled data using a multi-stage fine-tuning strategy. Extensive experimental results demonstrate that SemiRaman achieves superior performance on both Bacteria-7 and Bacteria-14, reaching 89.2 % and 92.0 % accuracy and corresponding MF1-score of 89.1 % and 91.0 % with only 5 % and 10 % labeled data, respectively. Furthermore, compared with other baseline methods, SemiRaman exhibits outstanding performance in semi-supervised settings, achieving stable and reliable performance on complex pathogenic bacterial Raman spectroscopy datasets-even with extremely limited labeled data and in the presence of strong noise and high species diversity. In short, SemiRaman provides a promising and cost-effective solution for the rapid and accurate identification of hazardous pathogenic bacteria in food safety, environmental monitoring, and public health.