Deep-learning-based gas identification by time-variant illumination of a single micro-LED-embedded gas sensor

Area of Science:

• Chemical Sensors
• Artificial Intelligence
• Optoelectronics

Background:

• Electronic nose (e-nose) technology faces challenges with chemoresistive sensor cross-reactivity.
• Existing e-nose systems require complex sensor arrays for gas identification.
• Smart factories and personal health monitoring demand efficient gas sensing solutions.

Purpose of the Study:

• To develop a novel sensing strategy for selective gas identification and quantification.
• To overcome the cross-reactivity limitations of traditional chemoresistive sensors.
• To enhance the efficiency of e-nose technology in terms of cost, space, and power.

Main Methods:

• A micro-LED (μLED)-embedded photoactivated (μLP) gas sensor was designed.
• Time-variant illumination from the μLED generated forced transient sensor responses.
• A deep neural network analyzed transient signals for gas detection and concentration estimation.

Main Results:

• The single μLP gas sensor achieved high classification accuracy (~96.99%) for toxic gases.
• Quantification accuracy (mean absolute percentage error ~ 31.99%) was demonstrated for various gases.
• The sensor system operated with a low power consumption of 0.53 mW.

Conclusions:

• The proposed μLP gas sensor with time-variant illumination offers a novel solution for selective gas sensing.
• Deep neural network analysis enables accurate gas identification and concentration estimation from transient signals.
• This approach significantly improves e-nose efficiency, reducing cost, space, and power requirements.