A qualitative analysis method for UV Raman spectra of domain - adaptive deep convolutional neural network

Abstract Raman spectroscopy technology is widely employed in materials science, chemistry, medicine, and environmental monitoring due to its non-destructive nature and high sensitivity. However, significant discrepancies in the obtained Raman spectra for the same sample across different spectrometers present challenges for qualitative analysis. To mitigate these challenges, this study introduces a domain-adaptive deep convolutional neural network (DADCNN) to facilitate the effective transfer of spectral features between instruments. Specifically, deep convolutional neural networks successfully capture complex patterns and subtle differences in Raman spectra, enhancing the recognition of spectral features. Furthermore, a domain adaptation function is introduced to align the feature distributions of the source and target domains, thereby reducing spectral feature deviations caused by instrument discrepancies. Experimental results demonstrate that the proposed method substantially improves the accuracy of qualitative spectral data analysis under few-sample conditions, achieving a prediction accuracy of over 90%, surpassing traditional methods and other transfer learning models. This study offers novel insights and solutions for the sharing and application of Raman spectra across various instruments.