Qualitative identification of passive Fourier transform infrared spectra via an attention mechanism based deep learning approach

Passive Fourier transform infrared spectroscopy, used to detect chemical pollutants in the air, works with extremely weak signals with complex and varying background interference. This significantly challenges gas identification in terms of precision. Consequently, limited research progress has been made in this area. To address this issue, this study proposes a model that leverages the Transformer, a self-attention-based neural network, and a coattention mechanism for gas identification. The architecture of this model facilitates joint feature learning and fusion, rendering the prediction performance robust to background interference. Extensive experiments demonstrate its significant improvements and feasibility, underlining the potential application in hazardous gas warning systems.