Optical Gas Sensing with Liquid Crystal Droplets and Convolutional Neural Networks

Area of Science:

• Materials Science
• Chemical Sensing
• Artificial Intelligence

Background:

• Liquid crystal (LC)-based materials offer potential for developing rapid, miniaturized, and low-cost gas sensor devices.
• Hybrid gel films with LC droplets exhibit characteristic optical texture variations in response to volatile organic compounds (VOCs) due to molecular orientational transitions.

Purpose of the Study:

• To investigate the application of deep convolutional neural networks (CNNs) for analyzing optical texture dynamics in LC droplets exposed to various VOCs.
• To develop a pattern recognition system capable of identifying and quantifying VOCs using LC droplet responses.

Main Methods:

• Video recording of LC droplet responses to VOCs using polarized optical microscopy (POM).
• Utilizing deep convolutional neural networks (CNNs) to extract features from recorded optical textures.
• Training CNN classification models to recognize specific VOCs and regression models to quantify VOC concentrations.

Main Results:

• Individual LC droplets, analyzed by CNNs, demonstrated the ability to recognize 11 different VOCs with high accuracy (F1-score > 93%).
• The optical texture variation patterns within a single droplet correlated with VOC concentration changes, as confirmed by regression analysis for acetone (mean absolute error < 0.25% v/v).
• The study also analyzed the influence of droplet diameter on sensing performance.

Conclusions:

• CNN-based analysis of LC droplet optical textures is a highly effective method for sensitive and selective VOC detection.
• This approach enables the development of advanced, individual-droplet-based gas sensors with potential for miniaturization and cost reduction.