A Dual-Model Framework with Gramian Angular Field and Spatio-Temporal Attention for Rapid Gas Identification and...

Area of Science:

• Analytical Chemistry
• Artificial Intelligence
• Sensor Technology

Background:

• Accurate gas identification and concentration prediction are vital for industrial safety, medical diagnostics, and environmental monitoring.
• Signal distortion and feature loss in complex environments hinder electronic nose system performance.
• Existing methods struggle with data heterogeneity and robustness in real-world conditions.

Purpose of the Study:

• To develop a robust dual-model framework for electronic nose systems to enhance gas identification and concentration prediction.
• To address challenges of signal distortion, feature loss, and data distribution heterogeneity.
• To improve the accuracy and response speed of gas sensing.

Main Methods:

• A gas classification model utilizing composite Gramian Angular Field and Convolutional Neural Network (CNN).
• A gas concentration prediction model integrating multi-branch attention, CNN, and bidirectional Gated Recurrent Unit (GRU).
• A cascaded identification-prediction scheme to mitigate data heterogeneity and improve robustness.

Main Results:

• The classification model achieved 100% identification accuracy using initial response data.
• The prediction model attained R² > 0.99 for most target gases.
• The framework demonstrated strong adaptability to low concentrations, varying humidity, and gas mixtures.

Conclusions:

• The proposed dual-model framework offers an efficient and robust solution for rapid qualitative and quantitative gas analysis in electronic nose systems.
• The system effectively handles complex environmental conditions and supports both single-label and multi-label tasks.
• This advancement significantly improves the reliability and applicability of electronic nose technology.