Selectivity Enhancement in Electronic Nose Based on an Optimized DQN.

Area of Science:

• Materials Science
• Chemical Sensing
• Artificial Intelligence

Background:

• Metal oxide gas sensors offer potential for environmental monitoring but often suffer from poor selectivity.
• Exploiting transient sensor information is crucial for improving the performance of gas sensing systems.

Purpose of the Study:

• To enhance the selectivity of metal oxide gas sensors using flow modulation and deep reinforcement learning.
• To develop an active perception strategy for optimizing gas classification in real-time.

Main Methods:

• A flow modulation method was employed, dynamically adjusting carrier gas flow to optimize transient sensor signals.
• A Deep Q-Network (DQN) was utilized for online optimization of flow modulation, learning an action policy to maximize rewards (minimize classification error).
• A Convolutional Neural Network (CNN) was trained to predict sample class and reward based on sensor observations and actions.

Main Results:

• The proposed DQN-based flow modulation achieved an average gas classification accuracy of 88%, significantly outperforming traditional Principal Component Analysis (PCA) which averaged 69.6%.
• Specific gas accuracies included sesame oil 100%, lactic acid 80%, acetaldehyde 80%, acetic acid 80%, and ethyl acetate 100%.
• The DQN method demonstrated higher recognition accuracy, improved recognition speed, and reduced training/testing costs compared to PCA.

Conclusions:

• Flow modulation combined with DQN offers a powerful strategy for enhancing metal oxide gas sensor selectivity and classification accuracy.
• The active perception approach shows adaptability, particularly under conditions with limited labeled data.
• This intelligent sensing system provides a promising direction for real-time, high-performance gas detection.