Multichannel Hierarchical Analysis of Time-Resolved Hyperspectral Data for Advanced Colorimetric E-Nose

Area of Science:

• Chemical sensing
• Spectroscopy
• Machine learning

Background:

• Colorimetric sensor-based electronic noses are used for gas molecule discrimination in health and environmental monitoring.
• Conventional systems using RGB sensors have limitations in capturing complete spectral responses, hindering accurate chemical identification, especially for similar functional groups.

Purpose of the Study:

• To introduce a novel time-resolved hyperspectral (TRH) dataset for colorimetric array sensors.
• To enable hierarchical analysis of multichannel 2D spectrograms using a convolution neural network (CNN).
• To demonstrate the superior classification performance of TRH data compared to RGB data.

Main Methods:

• Developed a TRH dataset with 1D spatial, 1D spectral, and 1D temporal axes.
• Utilized a CNN for hierarchical analysis of multichannel 2D spectrograms.
• Measured and trained CNN models using TRH and RGB sensor systems at different relative humidity (RH) levels.

Main Results:

• The TRH model achieved 97.5% classification accuracy for RH concentration, while the RGB model achieved 72.5% under identical conditions.
• Demonstrated the detection of various functional volatile gases using the TRH system through experimental and simulation approaches.
• Observed distinct spectral features in the TRH system corresponding to changes in substance concentration.

Conclusions:

• TRH data significantly enhances the performance of machine learning analysis in electronic nose applications.
• The proposed TRH system offers a more accurate and robust method for chemical sensing and identification.
• TRH data provides distinct spectral signatures for identifying different volatile compounds.