Abstract
Chemical vapor sensors are essential for various fields, including medical diagnostics and environmental monitoring. Notably, the identification of components in unknown gas mixtures has great potential for noninvasive diagnosis of diseases such as lung cancer. However, current gas identification techniques, despite the development of electronic nose-based sensor platforms, still lack sufficient classification accuracy for mixed gases. In our previous study, we introduced multichannel hierarchical analysis using a time-resolved hyperspectral system to address the spectral ambiguity of conventional RGB sensor-based colorimetric e-noses. Here, we demonstrate the identification of mixed gas components through time-resolved line hyperspectral measurements with an eight-colorimetric sensor array that uses genetically engineered M13 bacteriophages as gas-selective colorimetric sensors. The time-dependent spectral variations induced by mixed gas in the different colorimetric sensors are converted into a hyperspectral three-dimensional (3D) data cube. For efficient machine learning classification, the data cube was converted into a multichannel spectrogram by applying a novel data processing method, including dimensionality reduction and a block average filter to reduce high-dimensional complexity and improve the signal-to-noise ratio. A convolution filter was then used for hierarchical analysis of the multichannel spectrogram, effectively capturing the complex gas-induced spectral patterns and temporal dynamics. Our study demonstrates a classification accuracy of 93.9% for pure and mixed gases of acetone, ethanol, and xylene at a low concentration of 2 ppm.