In industrial production and scientific research practice, the presence and transportation of hazardous chemicals are accompanied by many potential safety hazards and environmental risks. Therefore, it is crucial to adopt effective classification testing for hazardous chemicals. Raman spectroscopy technology has the advantages of non-contact, non-destructive, and fast detection. Moreover, Raman spectroscopy can accurately obtain the fingerprint spectral information of substances and has unique advantages for the identification and detection of hazardous chemicals. Owing to the significant errors concomitant with manual Raman spectroscopy analysis, combining convolutional neural networks in deep learning provides new ideas and methods for the analysis and processing of spectral data. By combining Raman spectroscopy with deep learning algorithms and introducing attention mechanism into convolutional neural network, the proposed algorithm achieves an accuracy of 99.47% in the classification of 500 hazardous chemicals.