Dual-branch transfer learning in Raman spectroscopy for bacterial quantitative analysis

Accurate quantification of bacteria is critical for ensuring food safety, advancing biomedical research, and a range of other pressing concerns. Raman spectroscopy is a popular technique for quantitative analysis due to its benefits of being fast, non-destructive, and highly sensitive. However, the accuracy of the transfer model is often limited by factors such as differences in equipment and environmental noise, which limits the popularization of Raman spectroscopy. In this paper, we propose an approach that overcomes this challenge by introducing a dual branch network based on Continuous Wavelet Transform (CWT) for model transfer. Our model comprises dual branches that perform distinct tasks. The spectral learning branch is responsible for extracting features from the spectral domain. The time-frequency map learning branch employs CNNs for extracting the multi-scale information-rich features. The proposed method is used for the quantitative analysis of Escherichia coli. The proposed approach significantly outperforms traditional methods in improving prediction accuracy. It offers a much-needed solution to the long-standing challenge of Raman spectroscopy in the field of bacterial quantitative analysis. With our approach, we can expect to see Raman spectroscopy more widely adopted in the future.