Quantitative analysis of pathogens is crucial for the prevention and treatment of infectious diseases. Compared to traditional microbiological identification methods, Raman spectroscopy offers advantages such as speed, non-destructiveness, and high sensitivity. However, it faces limitations including long analysis times and the requirement for specialized expertise. To address these challenges, this paper proposes a method combining wavelet transform and Transformer models for the precise detection of pathogens. The proposed method is validated on a publicly available Raman spectroscopy dataset of pathogens, with comparative analysis against Random Forest, VGG19, ResNet, and AlexNet algorithms. The results demonstrate that the spectral data processed with wavelet transform achieves a 3% accuracy improvement on the Transformer model compared to the raw data. Specifically, the accuracy reached 95.21% for the classification of 30 types of pathogens and 99.2% for the classification of 8 types of antibiotics. The method outperforms the comparative algorithms in terms of classification accuracy, and also exhibits high recall and F1 scores. This study enhances the efficiency and accuracy of rapid bacterial infection diagnostics, providing a novel tool for biomedical detection research.